Oil inflows and housing market fluctuations in an oil-exporting country: Evidence from Iran

Journal of Housing Economics 30 (2015) 59–76

Contents lists available at ScienceDirect

Journal of Housing Economics journal homepage: www.elsevier.com/locate/jhe

Oil inflows and housing market fluctuations in an oil-exporting country: Evidence from IranR Nasser Khiabani a,b,∗ a b

Department of Economics, Allameh Tabataba’i University, Shahid Beheshti,Tehran, Iran Department of Economics, Institute for Management and Planning Studies, Niyavaran, Tehran, Iran

a r t i c l e

i n f o

Article history: Received 27 April 2011 Revised 1 October 2015 Accepted 5 October 2015 Available online 19 October 2015 JEL classification: C32 C53 E52 E32 Keywords: Housing market fluctuations Oil price shocks Money shocks Bayesian Structural VAR Posterior model probability (PMP) Bayesian Monte Carlo integration method

a b s t r a c t In this study, I develop a structural macroeconomic model for Iran, an oil exporting country, highlighting the transmission channels of oil prices to the housing market. The model combines three blocks of macroeconomic interest consisting of money, goods and foreign exchange markets with the housing market. I identified and estimated the model using a Bayesian structural vector autoregressive framework. I also use posterior model probabilities to deal with uncertainty in the identification scheme and Bayesian Monte Carlo integration methods to obtain the correct posterior distribution for the structural parameters and to generate accurate confidence intervals for the impulse responses. The findings indicate that oil price shocks have a positive and persistent effect on housing activities. In contrast, money expansion has limited effect on housing market variables. Quantitatively, in the medium and long run, positive oil shocks explain about 28% of the variation in housing stocks and 21% of the variation in real housing prices. By contrast, money shocks explain 11% and 5% of the variation in housing stocks and housing prices, respectively © 2015 Elsevier Inc. All rights reserved.

1. Introduction In the last decade many oil exporting countries have experienced an extraordinarily sharp increase in housing prices, which is accompanied by a period of high oil inflows in these countries. This observation raises a number of questions which are potentially important for explaining the surge in housing prices in these countries. Does the sharp rise in oil prices over the last decade represent R An earlier version of the paper was presented in 3th annual conference on Iran’s economy, the University of Chicago, 15–17 October 2010. I am grateful to Hashem Pesaran, two anonymous referees and the Editor (Tom Davidoff)for insightful comments. ∗ Corresponding author at: Department of Economics, Allameh Tabataba’i University, Shahid Beheshti, Tehran, Iran E-mail address: [email protected], [email protected]

http://dx.doi.org/10.1016/j.jhe.2015.10.002 1051-1377/© 2015 Elsevier Inc. All rights reserved.

a key factor in boosting housing activities in these countries? If so, what are the transmission channels of this interaction? From a theoretical point of view, there is a substantial literature that investigates the transmission and propagation of the inflow of oil windfalls to the rest of economy. In this context, the Dutch disease is a well-known view that deals with ‘oil inflows problem’ and its implication for the traded and non-traded sectors in oil exporting countries. The main concentration of the Dutch disease is on large foreign exchange inflows that are result in higher real income and an appreciation in real exchange rates. More specifically, higher oil income and hence an appreciation in the real exchange rate, provide incentives to increase consumption and decrease the production of traded goods and vice versa for goods which are not traded internationally. See, for example,

60

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

Cordan (1982), Fardmanesh (1991), Van Wijnbergen (1984), Gylfason (2001), Caselli and Cunningham (2009) and Cologni and Manera (2013). However, oil inflows can also have important monetary effects, which will spill over to the rest of economy (Aoki and Edwards, 1983; Edwards, 1986). Based on this view, oil windfalls will typically result in the accumulation of foreign reserves. Assuming that this increase in the reserves is not sterilized, the monetary base will increase and an excess supply of money may develop. This increase may induce an impact on the price of non-tradable good and an appreciation in real exchange rates. While these theoretical considerations give us some tentative indications about the channels of oil inflows, they obviously do not clearly shed light on the transmission of oil inflows to housing activities. However, emphasizing the second view of the Dutch disease and taking into account the credit and financial channels of this aspect to the housing market, we may identify the transmission of oil inflows to housing markets. While considering these points, Oil inflows can have crucial credit effects, which will spill over to the real exchange rate, housing sector and real activities. Oil inflows, directly accumulate the foreign reserves and this causes the monetary base of the central bank to expand. Since credit rationing policies and the imposition of an aggregate ceiling on the stock of credit are conventional features of monetary policy in developing oil exporting countries,1 oil inflows may extend the credit ceiling of banking system. This eases access to credit resources by households and firms and lead to a significant expansion in the demand for traded and non-traded goods and makes housing as a quintessential non-tradable asset. Since the supply of non-traded goods and housing assets is more limited than that of traded goods, the upward pressure on their demand leads to an increase in the relative price of non-traded goods (i.e., a real exchange rate appreciation) and an increase in the relative price of housing assets.2 On the other hand, an increase in housing prices raises the value of the housing capital, which feeds into higher net worth for the household sector. This in turn, encourages them to borrow more to finance housing investment. In contrast to these linkages that have been much less focused in the major oil exporting countries, there are vast studies that have discussed the interaction between housing prices, monetary variables and the macro-economy, particularly in advanced economies where most of them use structural vector autoregressive or DSGE models. These models particularly focus on the role of housing channels in the monetary transmission mechanism and the role of housing wealth and collateral effects in linking credit and house prices. For example, Lastrapes (2002), Jarocinski and Smets (2008), Iacoviello (2002), and Elbourne (2008), Himmelberg et al. (2005) and Goodhard and Hoffman (2008) use a

1

For more details see Agenor and Montiel (2006). Note that since many developing oil exporting countries follow a fixed or a tight managed exchange rate policy, oil windfalls slow down the nominal exchange rate and consequently, the rate of domestic traded inflation. Therefore, it would be more likely to expect an appreciation in the real exchange rate resulting, partially, from the slowing down in the inflation of the domestic tradable goods and partially from higher prices of housing and other non-tradable goods. 2

structural VAR model to analyze the transmission of monetary policy shocks to housing variables. On the other hand, Aoki et al. (2002), Iacoviello (2005) and Iacoviello and Neri (2010), analyze the propagation of the shocks on housing activities in a DSGE framework. The general finding of these studies indicate that monetary policy shocks have an important role in explaining cyclical movements in the housing market. In this paper, I identify and quantify the interaction between oil prices and housing activities by paying attention to the above extended transmission mechanism of oil inflows to the housing market. The paper has several contributions to the existing literature. First of all, to my knowledge, this is the first paper that formally addresses the interaction of oil prices and housing activities via oil inflows channels in an oil exporting country. I concentrate on Iranian data since the Iranian housing market provides a particularly interesting case study. This is because, there have been large-scale increases in the price of owner occupied dwellings in recent years and these increases have occurred during oil price boom. Second, I develop a structural macroeconomic model for an oil exporting country to consider the main linkages of oil inflows with the housing market. The model can track the dynamic transmission of oil prices shocks to housing variables and address all the relevant questions that I have raised above. From an econometric point of view, the model is identified and estimated using a Bayesian structural vector autoregressive approach that combines the three blocks of macroeconomic interest consisting of money, goods and foreign exchange markets with the housing market. The identifying scheme of the model is achieved by imposing enough prior restrictions derived from the structural macroeconomic model developed in this paper. Since the theoretical identification scheme for structural shocks turns out to be a relatively large and over-identified structural VAR model, relying on a Bayesian approach can give precise estimation of structural coefficients. Furthermore it can produce error bands whose possible asymmetries are justifiably interpreted as informative about asymmetry in the posterior distribution of the impulse responses (see Sims and Zha (1999) and Waggoner and Zha (2000)). I examine the model using Iranian data over the period 1988:2–2013:4. The results indicate that the oil price shock has an important influence on housing prices and the housing stock. The findings indicate that oil price shocks have a significant, positive and persistent effect on housing activities. In contrast, credit expansion shocks have smaller and less significant effect on housing market variables. Housing prices and the housing stock increase in response to a positive credit shock, but the magnitude of these increments is noticeably smaller than that of responses to a positive oil price shock. The results also show that, in the medium and long run, positive oil shocks explain about 28% of the variation in housing stocks and 21% of the variation in real housing prices. By contrast, credit shocks explain 11% and 5% of the variation in housing stocks and housing prices, respectively. The remainder of the paper is organized as follows. In Section 2, I summarize the most important institutional events during the sample period. The theoretical economic model, a structural Bayesian VAR model and the

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

1.2

61

2.4

Real house price Real price of oil

1.0

2.0 1.6

0.6

1.2

0.4

0.8

0.2

0.4

0.0

0.0

-0.2

-0.4

-0.4

-0.8

log

log

0.8

-0.6

-1.2 88

90

92

94

96

98

00

02

04

06

08

10

12

Fig. 1. Real housing price versus real price of oil.

5.4

1.2

Real housing price Real exchange rate

5.2

1.0 0.8

4.8

0.6

4.6

0.4

4.4

0.2

4.2

0.0

4.0

-0.2

3.8

-0.4

log

log

5.0

3.6

-0.6 88

90

92

94

96

98

00

02

04

06

08

10

12

Fig. 2. Real housing price versus real exchange rate.

identification scheme are summarized in Section 3. Section 4 involves the estimation and identification process of the model. Section 5 presents simulation results based on the model. In this section, the robustness of the findings to alternative specifications is also considered. Finally, concluding remarks are presented in Section 6. 2. The Iran housing market in a birds-eye view Since the end of the Iran–Iraq war in the summer 1988, the Iranian economy has experienced several periods of rapid growth in its housing prices. Figs. 1 and 2 depict the behavior of real housing prices against the real price of oil and

real exchange rate respectively during 1988:2-2013:4. From these figures, we can identify three major price booms in the Iranian housing market over the mentioned period. The first two booms exhibit sharp “spikes” but the recent booms that started in 2000 has been much more extensive. The figures also show four slumps that took place in 1991, 1994, 1999 and 2005. According to these two figures, it seems that the housing price is pro-cyclical with the real price of oil, whereas it is counter-cyclical with the real exchange rate. It is also worth noting that the significance of this cyclicality increased strongly after 2000, when the price of oil began to increase. This first tentative result emphasizes the crucial role of oil

62

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76 Table 1 Macroeconomics and housing indicators (in percent changes). Period

1998–1991

1992–2000

2001–2013

Statistics variables

Mean

St. dev.

Mean

St.dev.

Mean

St. dev.

Real exchange rate GDP in constant prices Real world price of oil Money CPI Real housing price Real housing investment

1.1 9 1.6 20 15 –7 12

13 4 26 2.1 5.3 56 22

2 2.5 0.35 24 23 0.7 1.2

11 3 27 6 8 21 13

–3.3 3.7 11 24 17 9 10

17 4.2 34 4.5 6.6 30 16

price shocks in driving fluctuations in Iranian house prices (see also Table 1).3 During the above mentioned period, the usual explanation for part of the fluctuation in the housing market has been the rapid increase in money supply which led to high investment and/or speculative activities. Since interest rates have been set administratively during the sample period, the central bank could not set the interest rate by following a conventional monetary policy rule.4 During this period, the deposit and loan rates in the banking system changed little in comparison to a high and rising rate of inflation. On the other hand, considering the under-developed nature of the capital and bond markets, almost all the financing needs of the public and private sectors were met through the banking system.5 Therefore, the expansion of credit to the private and public sectors and the non-neutralized part of the country’s foreign exchange reserves, which depends on the country’s oil revenues, are among the most important driving forces behind money supply growth. Figs. 3 and 4 show the historical trend of the real price of housing and housing investment along with the real money supply, respectively. According to these figures, there has been a moderate correlation between the real money supply and housing investment prior to 1999 and a strong correlation since 2000. In the former period, the expansion of the money supply was mainly attributed to domestic credit growth, whereas in the latter period the expansion of money supply was mainly attributed to large foreign exchange inflows (which depended on the positive trend of the world price of oil). This suggests that, in the latter period, the money growth mainly originated from high oil prices and has been another source of rising residential investment in new dwellings. 3. The macroeconomics framework 3.1. The theoretical economic model This section considers a modified dynamic aggregate demand-supply framework that incorporates a housing sector and some important aspects of an oil-exporting 3 This evidence has also been seen in many members of OPEC, particularly since 2000. For example, there is strong co-movement between the house price and the world price of oil in Saudi Arabia, Kuwait, the UAE, Algeria and Qatar. 4 The Taylor rule is a very popular monetary rule that indicates that monetary authorities set short run interest rates in response to movements in output and inflation rates (see Taylor (1993)). 5 For more detail see Pesaran (2000).

economy. Oil price shocks and oil revenues play a major role in this economy. Financial and capital markets are underdeveloped, capital mobility is limited and the interest rates for bank liabilities are controlled. There are four fundamental markets in the economy: goods, money, foreign assets and housing. The goods market is specified with an emphasis on the role of oil revenues in the economy. In the money market, I adopt money demand in the usual way, but the money supply is characterized by the role of domestic credit and oil price shocks. In the foreign asset market, I consider two equations that have crucial roles in tracking external shocks, particularly oil price and risk premium shocks, on the domestic economy. These shocks can affect the domestic economy through their influences on the nominal interest rate and the real exchange rate. The former effect is based on imperfect capital mobility and the latter effect is based on purchasing power parity. The housing market is identified by modeling the demand and supply sides of the market. And finally, the inflation rate is determined by shocks driven from the money and goods markets and changes in both the real exchange rate and real housing prices. 3.1.1. Oil price process I assume that the real price of oil is an exogenous variable in response to instantaneous shocks in the economy. This assumption is justifiable, as Iran’s economy is small in magnitude and doesn’t have a large share of the world production of oil.

ot = εto

(1)

It is important to note that the exogeneity of oil price shock recently has being argued by some studies (see for example, Barsky and Kilian (2004), Kilian (2009)). In general there are two reasons to challenge this assumption. First, the existence of feedback effects between oil market and the macroeconomy. Second, the prices of oil are driven by distinct supply and demand shocks. For example, Kilian (2009) decomposes the changes in the price of crude oil to three shocks: the oil supply shocks, global GDP shocks and precautionary demand oil shocks. He shows that oil price shocks historically have been driven mainly by a combination of global aggregate demand and precautionary demand shocks rather than oil supply shocks. Although it is the beyond of the scope of this paper to provide a clear decomposition of oil prices shock into the three shocks, it is still possible to consider the impact of oil price shock, that can be originated from demand or supply shocks, on the Iran’s economy. This cannot be a strong limitation since Iran’s economy is small

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

8.6

63

1.2

Real housing price Real money supply

8.4

1.0 0.8

8.0

0.6

7.8

0.4

7.6

0.2

7.4

0.0

7.2

-0.2

7.0

-0.4

log

log

8.2

6.8

-0.6 88

90

92

94

96

98

00

02

04

06

08

10

12

Fig. 3. Real housing price versus real money.

8.4

6.0 Real housing investment Real money supply

5.5

8.0

5.0

7.8

4.5

7.6

4.0

7.4

3.5

7.2

3.0

7.0

2.5

6.8

log

log

8.2

2.0 88

90

92

94

96

98

00

02

04

06

08

10

12

Fig. 4. Real money versus real housing investment.

and we don’t expect to have a reverse causality from Iran’s economy to oil prices. 3.1.2. Inflation process In an open economy, the consumer price level is defined as a geometric average of the price of non-traded and traded goods:

pct = (1 − β2 ) pNt + β2 (et + p∗t )

(2)

where pct is the logarithm of the consumer price level, pN t is the logarithm of the price of non-traded goods, et is the logarithm of the nominal exchange rate and p∗t is the logarithm of the foreign price level. I also distinguish between house

prices and prices of other non-traded goods (that I call domestic output prices) and define pN t as a linear combination of the logarithm of house prices, pht , and domestic output prices, pt ,:

pNt = (1 − β3 ) pt + β3 pht .

(3)

By substituting (3) into (2) we obtain:

pct

= pt + β2 (et + p∗t − pt ) + β3 (1 − β2 )( pht − pt )

pct = pt + β2 ret + β3 (1 − β2 )r pht ,

(4) (4’)

where ret = et + p∗t − pt and r pht = pht − pt are the logarithms of the real exchange rate and the real house price level

64

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

respectively. Taking the differences of (4’) yields an inflation equation that can be written as:

 pct =  pt + β2 ret + β3 (1 − β2 )r pht − β2 ret−1 − β3 (1 − β2 )r pht−1 .

(5)

Note that we must now model the rate of domestic output price inflation, pt , the logarithm of the real exchange rate and also the real house price level on the right hand side of (5) to complete the inflation rate specification. At this stage, I discuss the determination of domestic output price inflation and leave the discussion the determinations of the real exchange rate and real house prices when I model the foreign exchange and housing markets. To model the rate of domestic output inflation, I assume  pt can be affected by shocks originating in the goods and money markets. Specifically, the rate of domestic output inflation is assumed to be related to contemporaneous shocks to output and money growth. Thus, normalizing in units of the money growth shock, we have:

 pt = β1 εty + εtm , εty

(6)

εtm

where and are real and money shocks, respectively. By substituting (6) into (5), we can eliminate the explicit consideration of the rate of domestic output inflation from the analysis. With these modifications (5) may now be rewritten as:

 pct = β1 εty + β2 ret + β3 (1 − β2 )r pht − β2 ret−1 − β3 (1 − β2 )r pht−1 + εtm

(7)

Note that εt and εtm can be identified in the goods and money markets, which are specified below. y

3.1.3. Goods market To specify behavior in the goods market, I assume that the logarithm of real output is given by:

yt = β4 εto + εty ;

mt = εtm = β7 εto + εtdc ; where

εtdc

β7 > 0,

(10)

is a credit shock.

3.1.5. Foreign exchange market I assume a general specification for the balance of payments that seems to be suitable for an oil-exporting country.

κ(it − it ∗ − ets + εta ) + (β8 ret + β9 ot + εtb ) = 0; β8 > 0, β9 > 0

(11)

where it ∗ is the foreign interest rate, εta is a risk premium shock, εtb is a trade balance shock, et is the logarithm of nominal exchange rate and the "s" superscript indicates the expected value next period. In (11) the first term determines the behavior of capital inflows whereas the second term determines the behavior of trade balance. The parameter κ denotes the degree of capital mobility assumed to be influenced by different measures of capital control or by prevailing institutional rules on internal financial markets, which can be modified to limit the speed of capital movements. I also assume that expectations on exchange rates are formed rationally:

ets = et + ut ,

(12)

where ut is a random prediction error. Rewriting (11) in terms of the domestic interest rate, by substituting (12) into (11) and using the definition of the real exchange rate (et = ret +  pt + i0 )6 :

it = i0 +  pt + (1 − β8 /κ)ret − (β9 /κ)ot − ret−1 + εtbop , (13)

β4 > 0,

(8)

where yt is the logarithm of real output and εt is a real shock which can be interpreted as a supply or demand shock. With this specification, monetary shocks are allowed to affect real output with a lag only, as is consistent with conventional views of the monetary transmission mechanism. y

3.1.4. Money market We assume a conventional money demand function. The demand for real money is contemporaneously correlated to interest rates, as well as to output shocks:

mt − pct = β5 εty + β 6 it + εtmd ;

the current value of money and other macroeconomic variables (Kim and Roubini 2000). Instead, I define money supply growth as monetary shocks which are correlated with oil price and credit shocks:

β5 > 0, β6 < 0,

(9)

where mt is the logarithm of money supply, it is the nominal interest rate and εtmd presents money demand shocks. As mentioned in Section 2, whilst conventional interest rate policies have not been the instigator of effective and significant monetary policy in Iran’s economy, changes in the level of aggregate liquidity directly depends on domestic credit channels and the non-neutralized part of the country’s foreign exchange reserves. In this regard, we are not able to define a conventional reaction function for the monetary authority, which sets the interest rate after observing

εtbop

it∗

+ ut − εta

+ εtb /κ ,

where = is the balance of payment shock. In Eq. (13) the interest rate is measured by the inflation rate, real exchange rate and real oil price augmented by a degree of capital mobility κ . As mentioned in Section 2, since the nominal interest rate in Iran is administratively determined and the mobility of capital is imperfect, a particular problem with data is that we have little confidence on the available interest rates reflecting market forces. Regarding this problem, I used the general definition of the interest rate in Eq. (13) to measure indirectly the effect of interest rate on system variables, particularly on housing variables. This relation has an important role in this study since it may propagate the effects of oil windfalls on the housing market via credit expansions and real exchange rate appreciations. I next model the behavior of the real exchange rate in order to complete the specification of the foreign market. Eq. (14) postulates that the logarithm of the real exchange 6 I assume the foreign price has a constant growth rate (− p∗t = i0 ). This assumption does not restrict the generality of our specification, because a non-constant foreign inflation rate is included as a shock in εtre in Eq. (14) specified below.

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

rate is contemporaneously correlated with oil price shocks, output shocks, monetary shocks and its own shock, εtre .

ret = β10 εto + β11 εty + β12 εtm + εtre ;

β10 < 0, β11 < 0, β12 > 0.

(14)

We expect that an expansionary monetary policy will lead to a contemporaneous real depreciation. Although theory does not impose particular priori restrictions on the sign of the output shocks in the real exchange rate equation, the expected negative effect of the output shock on the real exchange rate seems to be sensible. The inclusion of a real oil price term in the real exchange rate equation can be justified by the effect of the oil price on the traded and non-traded sectors in an oil-exporting country (see Garratt et al. (2003)). As discussed above, the real exchange rate channel has a crucial role in the Dutch disease context of transferring the oil price shock to the housing sector. 3.1.6. Housing market In this market, I concentrate on the behavior of three variables: housing stock, housing price and composite real construction cost. As shown in Miles (1994), the demand for housing can normally be derived from maximizing utility subject to an intertemporal budget constraint in a multiperiod or “life-cycle” approach:

ht = β13 εty + β14 (it −  pct ) + β15 r pht + εth ;

β13 > 0, β14 < 0, β15 < 0,

(15)

εth

where ht is the logarithm of housing stock Ht and is a housing demand shock. The anticipated theoretical signs of the partial derivative of the housing stock indicate that the housing stock is negatively related to the logarithm of real price of housing r pht and is positively related to the output shock. In addition, based on theoretical arguments alone, the logarithm of housing stock is also a negative function of the user cost of capital which is generally defined by the difference between nominal interest rate and inflation (Meen 1990, 2002).7,8 Solving (15) for r pht give us the inverted demand housing function:  ε y + β  (i −  pc ) + β  h + ε h ; r pht = β13 t t 14 t 15 t t  > 0, β  < 0, β  < 0. β13 14 15

(15 )

Empirical application of this equation depends on the nature of short term supply position. Under the assumption

of completely inelastic supply schedule for housing, one can consider the stock of housing as exogenous in (15’) and determine the adjustment of price toward the equilibrium value after a demand shock. On the other hand, under the assumption of an upward sloping supply schedule, the new investment can react to house prices in the short run. In this case we have to consider a housing supply equation along with (15’) to capture the feedback effect from housing prices to the investment in the new houses. A relevant economic theory for explaining this responsiveness of housing supply to housing prices is Tobin’s q theory of investment. Within this framework, the new construction is determined by the ratio of housing prices to construction costs. When the price of housing exceeds the full costs of developing and building in the construction sector, this then induces an increase in the quantity of dwellings supplied to the housing market (see for examples, Poterba (1984), Kenny (1999) and Madsen (2011) among others). Using the Tobin’s q approach of housing, we have equations:

HCt = f (RPth , CCt , It ) exp (εths ) Ht − Ht−1 = HCt + δ Ht−1 , where HCt is the new house completions, RPth is the real housing prices, CCt is the construction cost (including the land cost), δ is the rate of housing stock depreciation, It is the nominal interest rate where it can be approximated by it ∼ = log (1 + It ) and εths is a housing supply shock. The first equation says that the new completion of housing is positively related to real housing prices and negatively to the construction cost. In addition, since the nominal interest rate is well known as an informative indicator of the cost of financing investment for an average firm in the building sector, I also include the nominal interest rate in the equation.9 The second equation states that the change in the housing supply equals the new completions of housing minus a small fraction of the previous period’s stock. The steady state solution of the above two equations gives:

f (RPth , CCt , It )exp(εths ) = δ Ht−1 . Employing a log-linear approximation for the equation above and solving that for the real housing price, we can obtain the following form:

r pht = log (δ) +

β16 it + β17 cct + ht−1 + εths ;

β16 > 0, β17 > 0, 7

As pointed out in Meen (1990, 2002) the first order conditions of intertemporal utility maximization imply that the real price of housing is equal to the real imputed rental price of housing services divided by to the user cost of capital. Since the imputed rental price of housing is not directly observable in Iran, I use the housing stock and real income as a proxy for that. 8 Since the cost of owner-occupied housing services is measured by the user cost of housing, there are several other factors that should be considered in calculating the user cost of housing. Recently, the availability of collateralized credit and current and expected transaction costs are revealed to be the two important factors that are paid much attention by real estate literature (Grossman and Laroque 1990, Poterba and Sinai 2008 and Lustig and Van Nieuwerburgh 2005). Since this paper follows the link of macroeconomic effects to the housing market, it does not specifically concentrate on the housing characteristics of demand side in detail. It is also necessary to note that the reliable time series data for the housing collateralized credit and transaction cost are not available in Iranian 2economy. This limitation does not allow us to augment the user cost of housing with the above two mentioned variables.

65

(16)

where cct = log (CCt ) is the logarithm of real composite construction cost. The equation also implies that εths can be interpreted as any contemporaneous shock that affects r pht but is uncorrelated with construction cost shocks. In this regard, we are able to decompose construction cost shocks from other housing supply shocks. It is also worth noting that, considering land cost in cct may play an important role in the effectiveness of Tobin’s q model in the supply side of housing. If the scarcity of land becomes a serious obstacle for housing development, the positive effect of increased housing prices

9 The nominal, as opposed to the real, interest rate is used because financing costs are not related to discounting of a real income flow but are a direct expense (for more detail see Madsen 2011).

66

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

on new investment is likely to be quickly offset by the endogenous increasing of composite construction cost due to rising in land cost. Regarding this explanation, I complete the specification of the housing market with modeling the behavior of real composite construction cost. As shown in (17), the real composite construction cost is considered as function of the real output y shock εt and its own shock εtcc . Where the latter shock reflects building materials and the other related shocks. In addition, it is also possible to include the interest rate in (17) to capture the effect of cost of borrowing on the construction cost. The evaluation of this alternative choice model is done in Section 5.3.

cct = β18 εty + εtcc β18 > 0

(17)

The solution of structural model, (1) to (17), can be obtained by using (13) to eliminate explicit consideration of the interest rate from (9), (15) and (16) and after performing some straightforward algebra:10

ot = εto

 pct = γ21 ot + γ23 yt + γ25 ret + γ27 r pht + εtdc yt = γ31 ot + εty mt − pt = γ41 ot + +γ42  pct + γ43 yt + γ45 ret + εtA ret = γ51 ot + γ52  pt + γ53 y + γ57 r pht + εtre ht = γ61 ot + γ63 yt + γ65 ret + γ67 r pht + εtB r pht = γ71 ot + γ72  pct + γ75 ret + γ78 cct + εtC cct = γ81 ot + γ83 yt + εtcc

(18)

pt , ret , ht , r pht , cct } can be written in term of a vector representation of the simultaneous equation model

A0 zt = ν +

(19)

3.2. Bayesian structural method and identification

10

(20)

where A0 is the contemporaneous coefficient matrix in the structural form and Ai are matrices that contain coefficients of lagged values of zt . et is a structural disturbance vector. The  ) = 0; ∀s = 0 such properties of et are: E (et et ) = , E (et et+s that is a diagonal matrix for which the diagonal elements are the variances of the structural disturbances and the offdiagonal elements are zero. The estimation of the free parameters in A0 and can be obtained by maximizing over the free parameters A0 and based on a conventional likelihood function11 and the impulse responses can be derived by imposing just enough restrictions on A0 to make a one-to-one mapping from reduce form parameters to A0 and . However, this procedure is generally disrupted when the model is over-identified. Although for an over-identified model the maximum likelihood estimation of A0 and provides an algorithm for mapping the reduced form parameters estimates into the structural estimates A0 and , it is not true that this mapping converts the posterior distribution of the unrestricted parameters correctly into the restricted parameters. Sims and Zha (1999) suggest a new procedure for generating Monte Carlo draws from the Bayesian posterior for the parameters in (20). As in Sims and Zha (1999), (20) is reparameterized such that: 1

0 zt = ν +

p 

i zt−i +ηt

(21)

i=1

γ21 = β7 − β1 β4 ; γ23 = β1 ; γ25 = β2 ; γ27 = β3 (1−β2 ); γ31 = β4 ; γ41 = β6 (β9 /κ) − β4 β5 ; γ42 = β6 ; γ43 = β5 ; γ45 = β6 (1 + (β8 /κ); z = 1 + β1 β12 ; γ51 = (β10 − β11 β4 + β12 β1 β4 )/z; γ52 = β12 /z; γ53 = (β11 − β12 β1 )/z; γ57 = −β12 β3 (1 − β2 )/z; γ61 = β14 (β9 /κ) − β13 β4 ; γ63 = β13 ; γ65 = β14 (1 + (β8 /κ)); γ67 = β15 ; γ71 = β16 (β9 /κ); γ72 = β16 ; γ75 = β16 (1 + β8 /κ); γ78 = β17 ; γ81 = β18 β4 ; γ83 = β18 ; εtA = εtmd + β6 εtbop ; εtB = εth + β14 εtbop ;

18

Ai zt−i + et

i=1

where:

εtC = εths + β16 εtbop .

p 

The dynamic representation of the theoretical model) (for the eight variables zt = {ot ,  pct , yt , mt −

Since it is required to justify only zero restrictions on the contemporaneous correlations among the first stage innovations in order to identify the SVAR, the lagged terms r pht−1 , ret−1 and ht−1 are omitted from explicit consideration in Eqs. (5), (13) and (16).

where 1 0 = −1/2 A0 and ηt = −1/2 et so that var (ηt ) = I. The likelihood function of (21) is

|1 0 |T exp{−1/2trace(1 0 S(Bˆ)1 0 )}

(22)

or

|1 0 |T exp{−1/2trace(1 0 1 0 S(Bˆ)) − 1/2trace((B − Bˆ) X  X (B − Bˆ)1 0 1 0 )}

(23)

where Bˆ = (X  X )−1 X  Z, S(Bˆ) = (Z − X Bˆ) (Z − X Bˆ), and the  , . . . , z ), τ th rows of Z and X are given by zt and (1, zt−1 t−p respectively. Taking the prior as flat in B and 1 0 , and by integrating over B, the marginal posterior on 1 0 can be obtained by:

p(1 0 ) ∝ |1 0 |T −v exp[−1/2trace(1 0 S(Bˆ)1 0 )]

(24)

where v = np + 1. Using |1 0 |v as an improper prior or as a consequence of starting with a flat prior on the coefficients of i ; i = 1, .., p, in (21), then converting to a parameterization in terms of 1 0 and reduced form coefficients B, eliminates discrepancies between posterior modes and maximum likelihood estimates. These enable us to obtain the correct posterior distribution for structural parameters and to generate accurate 11 The likelihood function can be shown by L(B, ) = | |−T/2 exp[−1/2trace(S(B) −1 )] where B and are reduced form pa-

rameters and variance-covariance matrix, respectively. The mapping function between structural parameters and reduced form is obtained by

= A−1 A0−1 . 0

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

67

Table 2 Reduced form diagnostic tests for VAR (2).

AR(2) F AR(4) F ARCH(2) F ARCH(4) F Normality (2) χ 2 (2)

O

p

Y

m–p

Re

H

rph

cc

0.38 (0.68) 1.19 (0.32) 0.47 (0.75) 0.47 (0.75) 2.8 (0.23)

1.7 (0.19) 1.9 (0.11) 0.78 (0.53) 0.78 (0.53) 0.0001 (1.0)

1.6 (0.19) 1.2 (0.28) 1.9 (0.12) 1.9 (0.12) 4.1 (0.12)

1.2 (0.29) 2.2 (0.07) 0.26 (0.89) 0.26 (0.89) 0.93 (0.62)

0.2 (0.81) 0.73 (0.57) 0.67 (0.61) 0.67 (0.61) 0.51 (0.77)

1.2 (0.29) 1.9 (0.11) 0.70 (0.59) 0.7 (0.59) 2.8 (0.23)

1.3 (0.26) 1.7 (0.16) 0.11 (0.97) 0.11 (0.97) 4.1 (0.12)

2.2 (0.11) 2.4 (0.06) 0.45 (0.76) 0.45 (0.76) 0.96 (0.61)

Note: Marginal significance levels for statistics are in parentheses.

confidence intervals for the impulse response of the coefficients. The restrictions embodied in 1 0 are summarized as:

⎡γ 11 ⎢γ21 ⎢γ31 ⎢ ⎢γ41 1 0 = ⎢ ⎢γ51 ⎢γ61 ⎣ γ71 γ81

0

γ22

0

0

γ23 γ33 γ43 γ53 γ63

0

γ83

0

γ42 γ52 γ72

0

0 0 0

γ44 0 0 0 0

0

γ25 0

γ45 γ55 γ65 γ75 0

0 0 0 0 0

γ66 0 0

0

γ27 0 0

γ57 γ67 γ77 0

0 0 0 0 0 0

γ78 γ88

⎤ ⎥ ⎥ ⎥ ⎥ ⎥. ⎥ ⎥ ⎦

(25) (n=8)((n=8)+1)

=36 free parameters While a maximum of 2 in the system (25) makes it just-identified, the existence of 31 parameters in the model imply that the system is overidentified. Using these elements, the structural parameters (β1 , β2 , . . . , β18 ) can thus be written as:

β1 = γ23 ; β2 = γ25 ; β3 = γ27 /(1 − γ25 ); β4 = γ31 ; β5 = γ43 ; β6 = γ42 ; β7 = γ21 + γ23 γ31 ; β8 /κ = 1 − (γ45 /γ42 ); β9 /κ = −(γ41 + γ43 γ31 )/γ42 ; β10 = γ51 + (γ51 + 1)(γ52 γ25 ) + γ53 γ31 + γ23 γ52 γ31 /(1 − γ25 γ52 ); β11 = (γ53 + γ23 γ52 )/(1 − γ25 γ52 ); β12 = γ52 /(1 − γ25 γ52 ); β13 = γ63 ; β14 = γ65 γ42 /γ45 ; β15 = γ67 ; β16 = γ72 ; β17 = γ78 ; β18 = γ83 . (26) Note from (29) that we need only 26 of the 31 elements

γ ij to estimate the 18 coefficients β i . This results in imposing 5 more cross-equation restrictions among the γ ij elements:

γ57 = −γ52 γ27 ; γ61 = −γ63 γ31 + γ65 (γ41 + γ43 γ31 )/γ45 ; γ71 = γ72 (γ41 + γ43 γ31 )/γ42 ; γ75 = γ72 γ45 /γ42 ; γ81 = −γ83 γ31.

(27)

4. Bayesian structural model estimation I examine the above eight-variable quarterly VAR model of the Iran economy over the period 1988:2–2013:4. Furthermore to account for the shifts in the series, I include four

dummies D95, D01, D98 and D90 to take the effects of the import compression and foreign debt repayment constrain of Iran in 1995, the terrorist attacks to the USA in 2001, the financial crisis of South East Asia in 1998 and Iraq–Kuwait war in 1990, respectively.12 Before I estimate the structural coefficients of the model, I first, in order to specify the VAR model correctly, asses the unit root properties of the variables and then, determine the optimal lag length of the VAR. Using an Augmented Dickey–Fuller approach and Schwarz criteria for choosing the optimal lag, all of the variables are found to have unit roots.13 Since the true order of the VAR in level is unknown, I have resorted to the usual VAR order selection criteria. To determine the lag length, the maximum likelihood ratio test, Akaike and Schwarz criteria are used. In this regard, the maximum likelihood ratio admits the existence of four lags whereas, Akaike and Schwarz criteria reached their minimum at six and two lags, respectively. As each of these three criteria determines a different lag length, it is essential to check the whiteness of the VAR residuals to distinguish the optimal lag length. Choosing 2 and 4 lag lengths is generally supported by the usual diagnostic tests (results for 2 lag lengths are reported in Table 2). For other lag lengths using the conventional significance level of five percent, I found the evidence of serial correlation and heteroscedasticity in the residuals for some of the VAR equations. In order to save degrees of freedom in estimating the model, I have therefore chosen 2 lags for the subsequent investigation. The estimation procedure can be evaluated in two stages. In the first stage I rely on the maximization of concentrated likelihood (22) and the Maximum likelihood ratio test to check the validity of the set of over-identifying restrictions. The maximization of concentrated likelihood (22) for the coefficients in (25) and imposing 5 restrictions in (27) is obtained by using a numerical optimization procedure. To select optimal initial values for parameters in 1 0 , the simplex method is adopted. After setting up the initial values, the parameters are obtained using the BFGS method. The LR test re) − 31 + sults indicate that the chi-square statistic with 8(8+1 2 5 = 10 degree of freedom is 14.6, and the significance level 12 The dummy variables are defined as follows: D95 = 1 in the period 1995:1-1995:2, and 0 otherwise; D01 = 1 in the period 2001:3-2002:3, and 0 otherwise; D98 = 1 in the period 1998:1-1998:4, and 0 otherwise and D90 = 1 in the period 1990:1-1990:4, and 0 otherwise. 13 Due to space limitations, the result of unit root test for the variables is not reported here.

68

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76 Table 3 The estimation of parametersa . Coff.

Coff.

Coff.

Coff. Coff. a b

γ 21

γ 23

γ 25

γ 27

γ 31

γ 41

γ 42

γ 43

γ 45

γ 51

γ 52

γ 53

.046 (.01)

γ 57 b

–.08 (.03)

γ 61 b

.24 (.09)

.17 (.04)

.03 (.02)

–.01 (.01)

–1.1 (.2)

.15 (.06)

γ 75 b

.05 (.026)

γ 78

–.35 (.13)

.72 (.28)

–.4 (.17)

–.12

–.002

.04 (.01)

–.01 (.006)

–.001 (.004)

.02

2.8 (1.1)

.12

.7 (.3)

–.01

.35 (.15)

8.3 (1.2)

γ 63

γ 71 b

γ 72

γ 22

γ 55

γ 66

γ 77

γ 88

12 (5)

68 (6)

43 (6)

78 (4.3)

31 (4)

478 (37)

14 (2)

–.08

.24

.22

.03

.15

–1.1

.04

24 (2) β 8 /κ .95

.04

–.22

–.001

2.8

.7

.35

β 13

β2

β 14

β3

β 15

γ 44

γ 67

γ 11 β1

γ 33

γ 65

β4

β 16

β5

β 17

β6

β 18

β7

β 9 /κ .01

γ 81 b

γ 83

γ 11

β 10

β 11

β 12

–.25

–.55

.87

The numbers in parentheses are standard errors. The parameters are indirectly calculated from cross restriction equations in (27).

is 0.15. Therefore, the over-identifying restrictions are not rejected at one percent. In the second stage, taking into account that the overidentified restrictions in (25) and (27) cannot be rejected at the 1 percent level, I rely on a Bayesian Structural Vector autoregressive method to estimate 0 and derive impulse responses and error bands. While our model is relatively large and over-identified, a Bayesian Structural Vector autoregressive method, suggested by Sims and Zha (1999) and Waggoner and Zha (2000), can give a precise estimation of 1 0 . Furthermore, it can produce error bands whose possible asymmetries are justifiably interpreted as informative about asymmetry in the posterior distribution of the impulse responses. The maximization of the marginal posterior density for the free coefficients in 1 0 and the restrictions (27) can be obtained by taking a flat prior on B and and using the same numerical optimization procedure in stage 1. The parameter estimates of γ ij and the associated t-statistics are reported in Table 3. The estimates are plausible, and most of them are significant at the 5% and 10% levels. I use these estimates and (26) to derive the coefficients, (β1 , β2 , . . . , β18 ), which are also represented in Table 3. Most of the parameters are correctly signed and well determined. First of all, I concentrate on the interest rate Eq. (13). As pointed out earlier, the equation has a crucial role in propagating oil windfalls in the economy. The estimated coefficients of the real exchange rate and oil price in the equation are 1 − β8 /κ = (.05) and β 9 /κ = (.01) that both of them are statistically significant. This suggests that the degree of capital mobility in the economy, over the sample period, is low and oil windfalls via the monetary, (10), and real exchange rate, (14), channels can drive down the real interest rate. As we will see later, this also has an important role in propagating oil price shocks to housing market, since a decline in the real interest rate due to oil inflows can create an upward pressure on the demand for housing. The coefficients β 1 (= –.1), β 2 (=.24) and β 3 (=.22), respectively, measure the contemporaneous effects of the real output shock, real exchange rate and real price of housing y on inflation. The negative sign of β 1 suggests that εt may be interpreted as an aggregate supply shock that contemporaneously affects inflation. The estimated coefficient of β 2

(=.24) indicates that about 75% of variation in total inflation is explained by the non-tradable component of inflation. About 25% of variation in non-traded inflation is attributed to the housing inflation. The estimated coefficients of output shocks β 5 (=.15) and interest rate β 6 (= –1.1) suggest a plausible money demand relationship. The coefficient β 7 (=.04) is a reasonable estimate of the contemporaneous effect of the price of oil shock on money growth. The real exchange rate equation includes the coefficients β 10 (= –.25), β 11 (= –.55) and β 12 (= .87) in which the first two coefficients imply that the positive shocks of oil price and output lead to a contemporaneous real appreciation and the last one implies that monetary expansion leads to a contemporaneous real depreciation, as would be expected.14 The economic intuition for Iranian housing data is conveniently represented in the estimated parameters of the housing equations. The coefficients β 13 (=.036), β 14 (= –.22) and β 15 (= –.001), respectively, measure the contemporaneous effects of income shock, user cost of capital and real price of housing on the real stock of housing. The contemporaneous effect of real income on the housing demand is statistically significant. The real interest rate has the correct sign and has a significant effect on the housing stock. This result suggests that a reduction in the real interest rate, which may be originated from a positive oil price shock, increases the housing stock. This suggestion will be clearly supported in the next section in which I show the magnitude of an oil windfall shock on the housing stock, significantly increases over the medium to long run. Note also that the coefficient of the real price of housing is not significant at the 10 percent level. As discussed in Miles (1994) in an economy with binding quantitative restrictions imposed on borrowers, the stock of housing is no longer

14 It is interesting to compare the two above estimated coefficients β 7 (= –.05) and β 10 (= –.25) with the results in Kim and Roubini (2000). In contrast to our results, Kim and Roubini estimate a positive relation between the money supply and oil price shocks for many G8 countries, with the exception of UK and Canada that have been known as oil exporting countries. It is also interesting to note that while in our study a positive changes in oil prices has a negative, sizable and significant effect on the real exchange rate, Kim and Rouibini find a positive relation between oil prices and exchange rates (E/$) for non-oil exporting G8 countries and a negative relation for oil exporting G8 countries.

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

necessarily a decreasing function of the price of housing (β 15 ≥ 0). The coefficients β 16 (= 2.8) and β 17 (= .7) show the significant contemporaneous effects of the interest rate and of the real construction cost on the real price of housing. Since, from (13) the nominal interest rate is positively related to the inflation rate and negatively related to the real exchange rate, the positive response of housing prices to the nominal interest rate may capture the effect of inflation risk premia in asset markets and the effect of the appreciation of real exchange rate in foreign markets. The former effect may be interpreted under the view that suggests the housing market is generally perceived to provide a good hedge against high inflation (Kenny 1999 and Salo 1994). The latter effect may also be interpreted under the monetary aspect of the Dutch disease literature that points out that booms in asset prices is accompanied with a flow of oil revenues to oil exporting countries and a real appreciation in domestic currencies. And finally, the coefficient β 18 (=.35) shows the positive effect of output shocks on the real construction cost. 5. Simulations After identifying and estimating the SVAR, I estimated impulse response functions and variance decompositions for the eight variables in order to investigate the dynamic interactions among them. As mentioned in Section 3.2, whilst the posterior function (22) is not in the form of any standard pdf, in order to generate error bands for the impulse responses I use a version of the random walk Metropolis algorithm for Markov Chain Monte Carlo (MCMC). The algorithm uses the multivariate normal distribution for the jump distribution on changes in parameters in 1 0 . I first simulate 15000 draws using a diagonal covariance matrix with diagonal entries .000001 in the jump distribution. These draws are then used to estimate the posterior covariance matrix of parameters 1 0 and scale it by the factor to obtain an optimal covariance matrix for the jump distribution; see Gelman et al (2004). 5.1. Impulse responses The responses of selected variables to a onestandardized-innovation in the world price of oil and the money supply with .84 flat-prior probability bands are shown in Fig. 5. First consider a one-standard-deviation positive shock to the price of oil given in the first column of Fig. 5. These figures clearly show that a positive shock to oil prices significantly increase real money, causing an appreciation to the real exchange rate, increase real output, decrease inflation and finally increase housing prices, housing stock and construction costs. More specifically, the positive oil price shock increases real money by about 2% at 6 quarters. It also causes the real exchange rate to gradually begin to appreciate. It reaches its minimum point by –3% (the maximum appreciation point) at 5 quarters and after that it reverts to its baseline in 14 quarters. The inflation response to the shock is negative. It declines to below its original steady state level and reaches its minimum point (by about –.7%) at 2 quarters, then reverts to its baseline. Real Output in response to the oil price shock, increases gradually and reaches its maximum point (by about 1%) at 8 quarters. A

69

positive oil price shock leads to an expansion of the housing sector. Only one quarter after the original shock, the price of housing increases and reaches its maximum point by about 4%, then quickly falls along this path to its steady-state level. The stock of housing increases sluggishly and permanently in response to the shock. It starts its increase after 4-quarter lags and reaches its steady-state level in above 14 quarters. More interestingly, the increase in the housing stock is much smaller than the proportional increase in housing prices following a sudden increase in oil prices. The graph also shows that a shock to the price of oil induces a rise in the cost base of an average construction firm. Whereas the price of housing quickly increases following a positive shock to the price of oil, the construction cost starts its increases after a one-quarter lag and slowly continues until it reaches its long run equilibrium. It can be seen from the graph that it takes approximately 14 quarters before the equilibrium ratio of house prices to construction costs is restored. It is worth noting that the sluggish upward adjustment of construction costs indicates a crucial role in restoring the equilibrium ratio of house prices to construction costs.15 All responses are statistically significant. These results support the view that the transmission of a positive oil price shock via oil inflows, money expansions and exchange rate appreciations are the most important or most explored channels in deriving booms in housing market in a major oil exporting countries. The response of the selected variable to the monetary expansion is shown in the second column of Fig. 5. The impact of a money shock on the real output is not statistically significant. The inflation response to the money shock is positive, statistically significant and sizable in impact. Housing prices increase in response to a positive money shock, but only with a noticeably smaller magnitude when compared with the response to a positive oil price shock. These results are consistent with the view that credit expansion has an expansionary effect on house prices. However, a justifiable interpretation for these responses in Iran perhaps is that the high inflation rate or the high inflation uncertainty associated with credit expansion, since, the housing market is generally perceived to provide a good hedge against future inflation. On the other hand, from a financial point of view, the financial accelerator is another important channel that may propagate and amplify the effect of money shocks on the housing market. See, for example, Kiyotaki and Moore (1997), Iacoviello (2005) and Iacoviello and Nari (2010)16 among others. Since the debtor’s borrowing capacity is constrained by the value of his or her collateral assets, the effect of a 15 These roles might be attributed to several factors that are conventional in the housing supply side literature. First, since supply is rigid and highly inelastic over the short-run, a much greater supply response is forthcoming as firms in the construction sector gradually react to changes in the profitability of home-building activity. Secondly, since the supply is severely constrained due to a lack of available land for housing development, construction costs endogenously respond to increase in land prices. And finally, although, over the short run, increasing in housing prices, open up an enhanced scope for earning profit in the house building sector, over the medium to long run, construction costs adjust to ensure a normal level of profit to the average construction firm (Tobins’ q effect). 16 If agents are constrained by credit, the price of a collateralized asset directly interacts with the debt level and therefore interacts with investment and output. Such an interaction can, in theory, generate a financial multiplier that amplifies business cycle shocks.

70 N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

Fig. 5. Impulse responses of oil price and money supply shocks.

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

71

Fig. 6. Variance decomposition of housing prices.

Table 4 Forecast error variance decomposition. Oil price shocks

Real money Real exc. rate Real hou. pric. Housing stock

Money shocks

1 year

3 years

5 years

1 year

3 years

5 years

11% 15% 17% 4%

25% 30% 21% 28%

20% 22% 20% 28%

46% 8% 6% 5%

13% 6% 5% 11%

8% 5% 5% 10%

positive money shock on housing prices may be reinforced by increasing the value of housing collateral and the borrowing capacity of debtors. Another channel for following the effect of monetary expansion on the housing market is the real exchange rate channel. The impact of monetary expansion is a significant depreciation of the real exchange rate. However, after the initial depreciation impact, the real exchange rate starts to appreciate quite quickly. This in turn can amplify the effect of monetary expansion on housing prices.

5.2. Variance decompositions As a final point of concern, Table 4 shows the variance decompositions of the key variables to oil price and money shocks. The results indicate that positive oil price shocks explain a large fraction of the variation in real money, real exchange rate, real housing price and real housing stock at all forecast horizons, particularly, in medium and long run. For example, at a forecast horizon of 20 quarters, oil shocks contribute about 20% of the variance in real money, 22% of the variance in real exchange rates, 20% of the variance in housing prices and 28% of the variance in housing stocks. The results again support the hypothesis that oil price shocks bear significant responsibility for the variability in the housing market in Iran’s economy. This may give us a preliminarily sense of

the explanation for housing market fluctuations in other oilexporting market. By contrast, money shocks have a weak to moderate effect on the housing market. For example after five years, shocks to money, explain up to only 8% of the variance of real money, 5% of the variance in real exchange rates, 5% of the variance in housing prices and 10% of the variance in housing stocks. This result is in contrast to those of other economies, specially developed ones, in which monetary shocks have major role in explaining house price fluctuations. See, for example, Iacoviello 2002 and Goodhart and Hofmann 2008, among others. In this regard, Iacoviello 2002 shows that around 40% of the variance of house prices in France, around 12% in Sweden and around 20% in UK are attributable to monetary shocks. Fig. 6 also shows the share of all variables in the variability of housing prices. Apart from the oil price shocks that contribute an important part of the variance after one year, the real exchange rate and construction cost shocks are also important. The real exchange shock contributes about 15% of the variance throughout the first three years and the construction shock as a supply shock contributes about 14% of the variance after 3 years. However the most cause of variation in the house prices is attributable to its own shock. Although the share of this shock in explaining the variation decreases along the horizon, it preserves its importance after 5 years. This result indicates that housing price shocks have an important role in forming the expectation of house prices in the future. 5.3. Robustness checks As pointed out in Stock and Watson (1996, 2001), impulse response functions in SVARs can be quite sensitive to changes in lag length, sample period and identification restrictions. I check the robustness of our model in each of the above three ways. Firstly, with the same identification restrictions, the impulse responses for 2, 4 and 5 lags are depicted in Fig. 7. In contrast to the sensitivity of some VAR

72 N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

Fig. 7. Sensitivity of impulse responses to different lags length.

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

Fig. 8. Sensitivity of impulse responses to different sub-samples.

73

74

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

models to the lag length, the impulse responses to all variables have the same shape and very similar timing. Secondly, the sensitivity of the results with respect to the sample period is tested by estimating the model for 4 truncated samples. The sub-samples cover 1988:2-2013,1990:2-2013:4, 1992:4-2010:4, 1988:2-2002:4 and 1988:2-2006:4 periods.17 Fig. 8, regarding these sub-samples and the full sample shows the impulse responses of the system variables with respect to oil price and money supply shocks. The general patterns of the responses in the sub-samples are the same as in the full sample and we do not observe a significant difference in sign and timing of the responses over all samples. Finally, I examine the robustness of our results to changes in the identifying restrictions of the model. I started the examination with two alternative sensible identifying restrictions. Firstly, I allowed the money shock to enter directly into the output equation in (8). Secondly, since builders construct houses relatively quickly, the cost of financing house construction may be contemporaneously related to the nominal interest rate (Mishkin 2007). To examine this hypothesis, I entered the nominal interest rate in the construction cost equation. These two alternative modeling choices alter the restrictions on γ ij coefficients in (25) and cross restrictions (27). I illustrate these new identifications scheme as follow:

⎡γ 0 0 0 0 0 11 γ γ γ 0 γ 0 21 22 23 25 ⎢ ⎢γ31 γ32 γ33 0 γ 0 35 ⎢ 0 ⎢γ41 γ42 γ43 γ44 γ45

= ⎢γ 2 0 0 γ55 0 ⎢ 51 γ52 γ53 ⎢γ61 0 γ63 0 γ65 γ66 ⎣ γ71 γ72 0 0 γ75 0 γ81 0 γ83 0 0 0 γ35 = −γ25 γ32 γ57 = −γ52 γ27 ; γ61 = −γ63 γ31 + γ65 (γ41 + γ43 γ31 )/γ45 ; γ71 = γ72 (γ41 + γ43 γ31 )/γ42 −γ83 γ31 ; γ75 = γ72 γ45 /γ42 ; γ81 = −γ83 γ31 γ37 = −γ25 (γ27 /(1 − γ25 )γ32 . ⎡γ 0 0 0 0 0 11 0 γ25 0 ⎢γ21 γ22 γ23 ⎢γ31 0 γ 0 0 0 33 ⎢ 0 ⎢γ41 γ42 γ43 γ44 γ45 3 0 = ⎢ 0 γ55 0 ⎢γ51 γ52 γ53 ⎢γ61 0 γ 0 γ γ 63 65 66 ⎣ γ71 γ72 0 0 γ75 0 γ81 γ82 γ83 0 γ85 0 γ57 = −γ52 γ27 ; γ61 = −γ63 γ31 + γ65 (γ41 + γ43 γ31 )/γ45 ; γ71 = γ72 (γ41 + γ43 γ31 )/γ42 ; γ75 = γ72 γ45 /γ42 ; γ81 = −γ83 γ31 + γ82 (γ41 + γ43 γ31 )/γ45 ; γ85 = γ72 γ45 /γ42 .

0

γ27 γ37 0

γ57 γ67 γ77 0

⎤ 0 0 ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥, 0 ⎥ 0 ⎥ ⎦ γ78 γ88

methods use the rules of conditional probability to make inference about unknown models given known data. If Data is the data and there are k competing models, M1 , M2 and M3 presented by the matrix restriction schemes 1 0 , 2 0 and 3 0 , then the posterior model probability can be given by:

p(Data|Mk ) p(Mk ) p(Mk |Data) = 4 , k=1 p(Data|Mk ) p(Mk )

(28)

where the marginal likelihood of the model is defined as

p(Data|Mk ) =



p(Data|φk , Mk ) p(φk |Mk )dφk .

φ k is a parameter vector that refers to (5) identified with k 0 restrictions. p(Data|φ k , Mk ) and p(φ k |Mk ) are the likelihood function and the prior density function of φ k , respectively. In line with Garratt et al. (2007), I set a non-informative prior for p(Mk ) that is the same for all four SVAR models and use an asymptotic approximation to the marginal likelihood of form:

log p(data|Mk ) ∝ l −

K log (T ) 2

(29)

which was proposed by Schwarz (1978). Where l denotes the log of the likelihood function (22) evaluated at maximum likelihood estimates (MLE), K denotes the number of parameters in the model and T is the sample size. I calculate the posterior model probabilities in (28) by maximizing the likelihood function (22) based on the three restriction schemes k 0 , k = 1, 2, 3. The posterior model probabilities for 1 0 , 2 0 and 3 0 are .73, .2 and .1, respectively. These results indicate that our first identification schemes specified by the restrictions 1 0 are best supported by the data in comparison with the other restrictions k 0 , k = 2, 3. I also derive the impulse responses of the model for the two alternative identification schemes. The results of the responses turn out to be consistent with our earlier results and do not affect the qualitative nature of our results in general.18 6. Concluding remarks

0

γ27 0 0

γ57 γ67 γ77 0

⎤ 0 0 ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥, 0 ⎥ 0 ⎥ ⎦ γ78 γ88

To compare the above restriction schemes with our main restriction scheme in 1 0 , I utilized posterior model probability method introduced by Garratt et al. (2007). Bayesian 17 It is necessary to note that, the shorter sub-samples are not really practical because of the large number of parameters to estimate.

This study develops a structural macroeconomic model to consider the main linkages of oil inflows and housing activities in Iran, an oil exporting country. The model is identified and estimated by a Bayesian structural vector autoregressive approach that combines the three blocks of macroeconomic interest consisting of money, goods and foreign exchange markets with the housing market. The findings indicate that an oil price shock explains a substantial part of housing market fluctuations. A positive oil price shock, through oil inflows and monetary channels, significantly increases housing prices and the housing stock over time. While the response of housing prices to oil price shocks is immediate and stronger in magnitude, the increase in the housing stock is sluggish and much smaller than the 18 I also checked the robustness of the model to different definitions of some variables. For example, I substituted the multilateral real exchange rate (measured based on the weighted wholesale price index of trading partners and the consumer price index for the home country) for the bilateral real exchange rate with and M1 for M2. None of these robustness checks altered the patterns of the impulse responses.

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

proportional increase in housing prices. This suggests that the supply side of housing is rigid and highly inelastic over short run. Moreover, there are severely constraints in this side which can be related to, for example, a lack of available land for housing development. The results also show that the positive adjustment of construction cost in the response to oil prices shocks has an important role in restoring the equilibrium ratio of house prices to construction costs. While house prices immediately jump up to a new equilibrium, construction costs show a sluggish upward adjustment behavior in reaching equilibrium. Monetary expansion is another important shock in explaining a part of housing market fluctuations, particularly in the short run. Housing prices and the housing stock increase in response to a positive money supply shock, however only with a noticeably smaller magnitude comparing to the responses of these to a positive oil price shock. . The results indicate that, in medium to long run, positive oil shocks explain about 28% of the variation in housing stocks and 21% of the variation in real housing prices. By contrast, money supply shocks explain 11% and 5% of the variation in housing stocks and housing prices, respectively. Overall, these findings suggest that managing oil inflows by adapting some appropriate monetary and fiscal rules may help to smooth fluctuations in the housing market. This stabilization policy may be implemented by fiscal smoothing, establishing oil stabilization fund, saving fund or financial fund, and choosing an optimal monetary policy in react to oil price shocks. Appendix A The sample period starts at 1988:2 because the Central Bank of Iran (CBI) started to publish quarterly data of national accounts after 1988. The quarterly data are seasonally adjusted using X-12 ARIMA seasonal adjustment program. However, the results are not sensitive to seasonal adjustment, i.e. the multiplicative method and other seasonal adjustment programs such as X-11 additive or multiplicative programs yield similar results. All variables are transformed to log scale in the empirical analysis. Variable definitions and a brief description of the data are listed below. World price of oil deflated by the U.S.’s consumer ot : price index (1996=100). Money supply M2 deflated by Iran’s consumer mt : price index (CPI) of CBI. pt : Inflation rate measured by Iran’s CPI. Real Gross Domestic Product at 1996 constant yt : prices of CBI (Quarterly National Accounts). Bilateral real exchange rate vis-a-vis the U.S. dollar ret : (its’ increment leads to the depreciation of the Rial versus the U.S. dollar). The indicator was calculated by dividing the nominal exchange rate (based on the market rate) to Iranian CPI. Housing stock that was computed using data on ht : housing completions published by the CBI. The measure was calculated using the perpetual inventory methodology by assuming a constant annual rate of housing depletion of 5% per annum. Housing price deflated by Iran’s CPI. Housing prices rpht : were computed as a weighted average of housing

75

prices at 33 large cities. The weights used in the calculation are based on the number of houses in the each of the 33 cites. The weight attached to the each of house prices is calculated by dividing the number of houses in the each of cities to the total number of houses in all cities. The housing price in each city, taken from Housing and Urban Development of Iran and Center of Iranian Statistics, is constructed based on housing transactions as an average value of housing at square meter. Construction cost index was calculated by weightcct : ing together the housing building cost index published by CBI and an index of the average price of land per housing unit calculated based on the average land prices of the 33 large cities discussed above. To do so, I obtained the average share of land cost in housing prices (SL ) by dividing the calculated land prices on the previous estimated average housing prices. I also estimated the share of housing building cost in the total cost by using the housing survey data published by the CBI. Since the share of housing building is estimated to account to no more than 63% of the total cost, I allocated the weight WL =.63/(SL +.63) to the housing building index and (1 – WL ) to the land cost index. The composite construction cost index is deflated by Iran’s CPI. Iran’s Consumer Price Index by CBI. pct : References Agenor, P.R., Montiel, P.J., 2006. Credit market imperfections and the monetary transmission mechanism Part I: Fixed exchange rates. The School of Economics Discussion Paper. Series 0628, Economics, The University of Manchester. Barsky, R.B., Kilian, L., 2004. Oil and the macroeconomy since the 1970s. J. Econ. Perspect. 18 (4), 115–134. Aoki, K., Proudman, J., Vlieghe, G., 2002. Houses as collateral: has the link between house prices and consumption in the UK changed. Policy Rev.: Financ. Innov. Monet. Transm. 8, 163–178. Caselli, F., Cunningham, T., 2009. Leader behaviour and the natural resource curse. Oxford Econ. Pap. 61 (4) 628.650. Cologni, A., Manera, M., 2013. Exogenous oil shock, fiscal policy and sector reallocations in oil producing countries. Energ. Econ. 35 (c), 42–57. Corden, W.M., 1982. Booming sector and Dutch disease economic: A survey. Australian National University Working Papers 79 (November). Edwards, S., 1986. A commodity export boom and the real exchange rate: The money-inflation link. In: Neary and van Wijnbergen (Ed.), Natural Resources and Macroeconomy. the MIT Press, pp. 229–248. Edwards, S., Aoki, M., 1983. Oil exports and Dutch-disease: A dynamic analysis. Resour. Energ. 5, 219–242. Elbourne, A., 2008. The UK housing market and the monetary policy transmission mechanism: An SVAR approach. J. Hous. Econ. 17, 65–87. Fardmanesh, M., 1991. Terms of trade shocks and structural adjustment in a small open economy: Dutch disease and oil price increase. J. Dev. Econ. 34, 339–353. Garratt, A., Koop, G., Mise, E., Vahey, S.P. Real-time prediction with monetary aggregates in the presence of model uncertainty, Birkbeck Working Paper 2007; 0714, available at http://www.ems.bbk.ac.uk/research/wp. Garratt, A., Lee, K., Pesaran, M.H., Shin, Y., 2003. A structural cointegrating macroeconomic model of the UK. Econ. J. 113 (487), 412–455. Gelman, A., Carlin, J.B., Stern, H.S., Rubin, D.B., 2004. Bayesian Data Analysis, 2nd edition Chapman & Hall. Goodhard, C., Hofmann, B., 2008. House prices, money, credit, and the macroeconomy. Oxford Rev. Econ. Policy 24 (1), 180–205. Grossman, S.J., Laroque, G., 1990. Asset pricing and optimal portfolio choice in the presence of illiquid durable consumption goods. Econom.: J. Econom. Soc. 57 (1), 25–51. Gylfason, T., 2001. Natural resources, education, and economic development. Eur. Econ. Rev. 45, 847–859.

76

N. Khiabani / Journal of Housing Economics 30 (2015) 59–76

Himmelberg, C., Mayer, C., Sinnai, T., 2005. Assessing high house prices: bubbles, fundamentals and misconceptions. J.Econ. Perspect. 19, 67–92. Iacoviello, M., 2002. House Prices and Business Cycles in Europe: a VAR Analysis. Boston College Working Paper, p. 540. Iacoviello, M, 2005. House prices, borrowing constraints and monetary policy in the business cycle. Am. Econ. Rev. 95 (3), 739–764. Iacoviello, M., Neri, S., 2010. Housing Markets Spillovers: Evidence from an Estimated DSGE Model. Am. Econ. J.: Macroecon., Am. Econ. Assoc. 2 (2), 125–164. Jarocinski, M., Smets, F., 2008. House Prices and the Stance of Monetary Policy. Federal Reserve Bank of St Louis Review, pp. 339–370. Kenny, G., 1999. Modelling the demand and supply sides of the housing market: evidence from Ireland. Econ. Model. 16, 389–409. Kim, S., Roubini, N., 2000. Exchange rate anomalies in the industrial countries: A solution with a structural VAR approach. J. Monet. Econ. 45, 561– 586. Kiyotaki, N., Moore, N., 1997. Credit cycles. J. Polit. Econ. 105 (2) 211.248. Lastrapes, W.D., 2002. The real price of housing and money supply shocks: Time series evidence and theoretical simulations. J.Hous. Econ. 11, 40– 74. Kilian, L, 2009. Not all oil price shocks are alike: disentangling demand and supply shocks in the crude oil market. Am. Econ. Rev. 99 (3), 1053– 1069. Lustig, H., Van Nieuwerburgh, S., 2005. Housing collateral, consumption insurance, and risk premia: An empirical perspective. J. Financ. 60 (3), 1167–1219. Madsen, J. A q Model of House Prices. Discussion Paper, Department of Economics, Monash University 2011; 03. Meen, G.P., 1990. The removal of mortgage market constraints and implications for econometrics modelling of UK house prices. Oxford Bull. Econ.Stat. 52 (1), 1–24.

Meen, G., 2002. The time series behavior of house prices: a transatlantic divide. J. Hous. Econ. 11, 1–23. Mishkin, F.S. Housing and the monetary transmission mechanism. NBER Working Paper 2007; No. 13518. Miles, D., 1994. Housing, Financial Markets and the Wider Economy. John Wiley and Sons. Pesaran, M.H., 2000. Economic trends and macroeconomic policies in postrevolutionary Iran. In: Alizadeh (Ed.), The Economy of Iran: Dilemmas of an Islamic State. I. B. Tauris, London, pp. 63–100. Poterba, J., 1984. A Tax subsidies to owner-occupied Housing: An Asset Market Approach. Q. J. Econ. 99, 729–752. Poterba, J., Sinai, T., 2008. Tax expenditures for owner-occupied housing: Deductions for property taxes and mortgage interest and the exclusion of imputed rental income. Am. Econ. Rev. 98 (2), 84–89. Salo, S., 1994. Modelling the Finnish housing market. Econ. Modell. 11, 250– 265. Schwarz, G., 1978. Estimating the dimension of a model. The Ann. Stat. 6 (2), 461–464. Sims, C.A., Zha, T., 1999. Error bands for impulse responses. Econometrica 67 (5), 1113–1156. Stock, J.H., Watson, M.W., 1996. Evidence on structural instability in macroeconomic time series relations. J. Bus. Econ. Stat. 14 (1), 11–30. Stock, J.H., Watson, M.W., 2001. Vector autoregressions. J. Econ. Perspect. 15 (4), 101–115. Taylor, J.B., 1993. Discretion versus Policy Rules in Practice. CarnegieRochester Conference Series on Public Policy, pp. 195–214. Van Wijnbergen, S., 1984. The Dutch disease: A disaster after all? Econ. J. 94, 41–55. Waggoner, D.F., Zha, T. Likelihood-preserving normalization in multiple equation models. Federal Reserve Bank of Atlanta Working Paper 2000; 8.