HOUSE PRICES IN AUSTRALIAN CAPITAL CITIES: A SUPPLY SIDE PROSPECTIVE

Transcription

1 HOUSE PRICES IN AUSTRALIAN CAPITAL CITIES: A SUPPLY SIDE PROSPECTIVE House prices and their dynamics have been at the forefront of recent academic debate which primarily focuses on the demand side factors like income, credit availability and interest rates. Little advancement has been made in understanding the potential supply side drivers of house prices. In this paper we investigate the supply side factors using a novel data set comprising of inner, middle, and outer regions of Australian capital cities. We contribute to the literature by introducing government revenue to the house price model as a mechanism to capture the value homeowner s attribute to the relocating or their continued residence within these regions. Allowing for time and regional fixed effects, and controlling for key demand side variables based on the existing literature, we find that dwellings completed is an important measure of housing stock in Australia which has subsequent bearing on house prices. Furthermore, the investigation reveals that land tax and stamp duty conveyance revenues both have a positive impact on house prices in Australian capital cities while the revenues generated from local municipalities exert downward pressure on the house prices. MATTHEW LARKIN DEAKIN UNIVERSITY Burwood Hwy, Melbourne, Victoria

2 Table of Contents Introduction... 2 Aim and Motivation... 2 Background and Contribution to Literature... 6 Empirical Strategy and Main Findings... 7 Data... 7 Methodology... 9 Results Fixed Effects Model Panel Vector Autoregression Robustness Checks Appendix Tables Taxes on Property References... 38

3 Introduction Aim and Motivation Government taxation revenue earned from the sale and/or occupation of property diverts a significant amount of funds from the housing sector. In this paper we introduce government taxation revenue into our model of house prices in Australian capital cities along side housing supply variables and a number of controls conventional to the current literature. There are two distinct sections of stare and local government property taxation that we are interested in. The first, taxes on immovable property, includes land taxes, municipality rates and other taxation. While revenue generated from taxes on financial and capital transactions is made up of financial institution transaction taxes, government borrowing guarantee levies, stamp duties on conveyances, and other stamp duties. The introduction of government taxation revenue allows us to proxy for the values of which owner occupants attribute to the continued ownership of a dwelling, and more specifically the land on which it resides. Housing, or more specifically the physical structure of a dwelling, does not hold a considerable amount of value. As an entity housing depreciates rather rapidly as construction materials age or the style in which it is designed becomes outdated. Once one controls for this quality adjustment and the combined effects of age- related depreciation and required maintenance, the dwelling its self holds considerably less value. As noted by Wilhelmsson, this impact of depreciation will differ depending on whether said property is well maintained or not (2008). Harding et. al. find, with the use of data from the American Housing Survey, house price appreciation is subtracted by between 2-3 percent per year once controlling for age- related depreciation (2007). In spite of this reality, we continue to observe real house price appreciation across our sample. Additionally we see that the more inner a region is, the more expensive house prices are. We intend to demonstrate that this is due to the value of which homeowners attribute to holding property in these locations. Table 1: Mean Real House Prices by Region Region Mean S.D. Inner Middle Outer Source 1: REIA There are a number of reasons why an individual or family would want to reside in these older residential areas. In their paper, Wilhelmsson indicates that older properties can experience appreciation even as the quality of the structure diminishes over time. They attribute this to a vintage effect (Wihelmsson 2008).In the Australian context there are a number of reasons why a property is more expensive in one location compared to another. Consumer preferences tend toward locations which have a greater availability of resources, vicinity to infrastructure and various other amenities. In addition households will also hold preferences over the construction of the properties they which to occupy including number of bathrooms, bedrooms as well as size of the various components that make

4 up a dwellings attributes. There is an extensive literature on hedonic house price indexes that accounts for various heterogeneous features which can an impact on the perceived quality of a dwelling, in particular the age and location (see Goodman and Thibodeau (1995) Goodman and Thibodeau (1998) Mills and Simenauer (1996) Stevenson (2004)). In more recent years there have been considerable developments in the spatial analysis of hedonic house prices to allow for various geographical features, finding that a dwellings location has a bearing on the final house price (studies include Liao and Wang (2012), and Helbich et al. (2014)) 1. In this paper we analyse a novel longitudinal panel of real house prices that incorporates eight Australian capital cities separated into inner, middle and outer regions. Due to a number of geographical features, such as vicinity to the central business district, as well as a number of demographic characteristics and varying access to amenities heterogeneity is apparent between these regions. With the implementation of our panel analysis, allowing for both regional and time fixed effects, we are able to elevate this heterogeneity problem to some extent. A further impediment of conducting sufficient house price analysis is the availability of adequate land data. Given that, as mentioned above, physical dwellings depreciate in value over time, while house prices continue to appreciate in the Australian context, we believe that it is imperative to find an adequate proxy for the underlying value that owners of dwellings give for the land on which a dwelling resides. Difficulty arises when attempting to obtain reliable data for these above occurrences. We attempt to circumnavigate this problem by incorporating government taxation revenue obtained from transactions of housing stock and occupation of dwellings into our model. Government revenue is generated in a number of different ways depending on which state a particular property resides. We examine several different state specific taxation revenues that are generated from the sale or occupation of a dwelling and the land on which it resides. In addition we incorporate a municipality revenue collected by local governments from owners of property in their relevant municipalities. In particular we are interested in three key government revenue variables. Land tax is collected by all states and the Australian Capital Territory 2, In general, land tax is to be paid by landowners with principle places of residence usually exempt from liability. Land tax is levied on landowners; while the ACT also incorporates all residential properties that are owned or rented by a trust or corporation. Since an owner occupant s principle place of residence is usually exempt from a land tax levy we can therefore interpret land tax as an opportunity cost of the annual possession of residential dwelling as an investment property. Conveyance revenue (known as stamp duty conveyance revenue) is an additional means of revenue for state governments. This duty is levied on written documents or instruments. Though this duty may be levied on a number of different transactions. we are interested is conveyance on real estate. Depending on the type of transaction some concessions may be afforded on the amount of conveyance payable. For instance a number of states and territories have relaxed the amount of duty payable for first home owners in recent years, thus providing greater incentive for new entrants into the housing market. A number of concessions noted by the Australian Taxation Office include: 1 For a in depth evaluation of the literature on hedonic house price indexes see Hill, RJ (2013). Hill indicates that 2 The exception being the Northern Territory who is yet to collect land tax revenue.

5 The Australian Capital Territory (ACT) does not levy duty on mortgages, hiring duty, leases and marketable securities. New South Wales (NSW) does not levy duty on leases, home loans to natural persons and the hire of goods. The Northern Territory does not levy stamp duty on mortgages, marketable securities, the grant and renewal of a lease and the hire of goods. Queensland does not levy duty on leases, marketable securities, credit card and credit business transactions, and the hire of goods. South Australia does not levy stamp duty on mortgages, leases, listed securities and the hire of goods. Tasmania does not levy duty on marketable securities, certain leases, mortgages and the hire of goods. Victoria does not levy duty on listed or unlisted marketable securities, certain leases, mortgages or the hire of goods. Western Australia does not levy duty on marketable securities, mortgages, hire of goods and leases. The data afforded to us demonstrates stamp duty revenue generated from property transactions in each state and territory. Using this information we can interpret stamp duty revenue as a transaction cost of purchasing a property. As the cost of relocating to or within a particular state increases a rational buyer will consider moving to another state where this transaction cost is lower. It is not, however, the only opportunity cost of relocating. There are significant burdens born families who have to relocate their dwelling including the removal and transfer time and cost, the necessity of finding new employment or schooling, even the time and effort spent adjusting to a new neighbourhood all contribute to the opportunity cost of relocating to a new property. This effect is amplified when the resident is not familiar with their new environment, which would most likely be the case if they were required to move some distance. Thus if a buyer observes that the opportunity cost of relocation is too great then they will remain in the same area and incorporate this proportion of the sunk cost into the value of the property.

6 Taxaion Revenue on Property: All States and Local Government 20,000 15,000 10,000 5,000 0 Land taxes Other Government borrowing guarantee levies Municipal rates Financial insituions transacions taxes Stamp duies on conveyances Other stamp duies Source: ABS As apparent in the above chart these three variables of interest incorporate the majority of taxation revenue generate by government from the ownership of transaction of property. We propose that the use of government revenue, particularly the three revenue types noted above as a variables of interest enables to get to the crux of the issue as to how home owners value the land, or location on which their dwelling resides. Rates are relatively constant over our ten year sample (DATA REFERECE), we control for the number of new dwellings created. Given that we have taken into consideration these two factors any correlation between house prices and changes in government revenue must be due to the value that occupant attribute to maintaining their current local. For instance, if land taxes were to increase in a particular location then the occupants would face the decision as to remain in said current dwelling or relocate to a dwelling in which the rate they pay is of a lower sum. Now the question becomes, does the home owner absorbed this taxation as a sunk cost of owning property or to they pass it on via an increase in their asset price. Given that housing stock is an incredibly illiquid asset we believe that a proportion of this revenue generated from taxation on housing can be attributed to house prices as home owners pass on a proportion of the housing taxation that they pay because the cost of paying this taxation is far less than the opportunity cost of relocating. Land is without a doubt the largest of the construction costs of erecting residential dwellings. We introduce government revenue as a proxy for land value, understanding that the amount of taxation on a property that an owner occupant is willing to pay has a direct relationship with the opportunity cost of relocating to an area where the cost of occupying a dwelling is less. To investigate this further we must first control for a number of variables that are identified as key determinants of house prices in the wider economic literature.

7 Having controlled for these variables we then introduce our housing stock variable. There are a number of papers that indicate they control for housing stock in Australia with dwellings sold. To the best of our knowledge this data is not available beyond We use dwellings completed which enables us to observe the change in housing stock over our sample. Finally, we introduce a vector of government revenue variables of which we are primarily interested in land taxation revenue, conveyance revenue and municipality rate revenue. Background and Contribution to Literature The price of housing in its simplest form is determined by forces of demand and supply. Naturally this has lead a number of economists to propose models which shed light on the plethora of variables that could potentially drive house prices. This, however, has led to a great deal of conflict in the literature depending on which model you preference. House price theory can be generally broken up into two separate theoretical sections. Asset pricing models which generally specify a user cost of housing. While present value models test the implicit relationship between house prices and rents. Neoclassical investment model focuses on the user cost of capital analysing the equilibrium value of imputed rental income accruing to home owners under various tax regimes. Authors of this approach include Poterba, JM (1992); Gyourko and Sinai (2004); Himmelberg, Mayer and Sinai (2005) to describe the homeowner s marginal cost of purchasing additional housing services. The latter is referenced heavily in Poterba s more recent work examining how federal income tax policy in the US affects homeownership for many households (2008). Empirically, drivers of house prices have been extensively studies with many papers suggesting that income, credit, interest rates and population all contribute to the demand for housing such as in Agnello and Schuknecht (2011). Due to the ongoing debate concerning the theoretical mechanisms at play in the determination of house prices we have decided to empirically model house prices in Australian capital cities in an attempt to shed light on the matter. One of the key limitations in modelling house prices is that the availability of land data is limited. To circumnavigate this issue we propose the introduction of a vector of government revenue variables. We believe that these variables capture some of the supply side effect on house prices through the inherent value owner occupant give to holding property in a particular region. The relationship between government revenue and house prices has yet to be considered in the Australian context and is yet to be fully understood in the wider community. Impacts of local government land leasing revenue on house price has been explored by Ding et al who utilise national and regional panel data analysis (2012) 3. They find a positive correlation between local government land leasing and house price, regional and time differences are apparent in their analysis. Regional differences are mainly reflected in differences in regional economic development, time differences are correlated with different government policies at different times. 3

8 The study of cross country and regional convergence has recently seen considerable advancements. Orsal and Orsal establish cross- country convergence with new cointegration test of in their recent paper (2013). Panel data estimates of cointegration relationships show a dependence on real house prices, real GDP per capita and real long- term interest rates. At a regional level convergence has also been studies in China, though demonstrate little evidence of convergence between regions (Zhang et al, 2014). Furthermore we must consider the elasticity of supply. If inelastic supply exists in the housing market then the impact of increased demand will be concentrated in price behaviour. If supply is highly inelastic or fixed in short term this could lead to rational bubbles (Stevenson and Young, 2014 p. 362). Glaeser et al. notes that developers have the ability to respond to increases in demand more rapidly, thus shortening the period of house price booms (2008). This is debatable, if developers have the ability to respond more quickly the alternative is also possible. Developers may instead choose to respond more slowly thus prolonging the period of labour demand. This would inturn restrict the availability of contemporaneous labour supply causing upward pressure on house prices through the cost of production. An assumption of imperfect elasticity in the housing market is supported by Poterba (1991), Kearl (1979), Schwab (1983), Topel and Rosen (1988) and Blackley (1999). At a regional level, Green et al. (2005) discuss the unusual properties of the supply curve in housing markets. Their findings indicate that supply elasticities vary greatly between metropolitan areas. Empirical Strategy and Main Findings To investigate our hypothesis that government revenue is a contributing factor to house prices in Australian capital cities, we employ a novel data set that makes use of house prices decomposed into three regions identified as inner, middle and outer. These regions approximately represent a 6km radius of the central business district (CBD), a 6-20km band from the CBD, between 20km and the outer limit of a cities boundary respectively. This panel data set provides the impetus from the novel empirical methodology that we utilise In this paper we make use of two complementary methodologies in order to exploit the rich data set that we have collected. Our findings suggest that land taxation revenue, conveyance revenue have a positive effect on changes in real house prices while municipality revenue has a negative impact on house prices in Australian capital cities. Furthermore, supporting findings of existing studies, both interest rates and dwellings complete both have a positive effect on house prices Data Our dependent variable, real house prices, is derived from data supplied from the Real Estate Institute of Australia (REIA). The REIA has made available to us median nominal house price data from 1988 to 2013 for all Australian capital cities. This has been further decomposed into three zones determined by their vicinity to the central business district. The inner zone demonstrates median house prices for

9 dwellings with three bedrooms or more in an approximate six kilometre radius of the CBD. 4 The middle zone is located in an approximate band between six and twenty kilometres from the CBD and the outer band from twenty kilometres to the outer boundary of the city. These 23 regions across Australia motivate the use of a panel structure to better understand the interaction of housing markets in Australia. Panel data analysis is at the forefront of house price analysis with a number of scholars examining the interaction of housing markets not only across regions. Simo- Kengne et al (2013) examine the endogenous relationship between house prices and consumption in a provincial- level panel vector autoregressive setting, Andrews & Calandra Sanchez (2011), and Adams & Fuss (2010) contribute further to this literature though across international borders (Sa et. al. 2014). This panel set up is of particular interest in the Australian context for a number for reasons. Firstly, due to the nature of Australian development the most populous areas are primarily located in the capital cites of the six states and two territories. Australia exhibits a relatively less dense population in comparison to many other industrialised nations analysed in panel studies, particularly those in Europe. This distance between capital cities is one very strong contributing factor that to their distinctly different housing marking characteristics, particularly contributing to the transaction costs involved if individuals wish to relocate between cities. We examine these transaction costs further through the introduction of government revenue as discussed below. The clustering of these metropolitan areas is indicative to cross sectional interaction between housing markets not only across states but also between the zones indented above. This interaction between the cities regions are most of interest, across each of the capital cities it can be seen that house prices are most expansive in the inner regions compared to the outer two. This could be due to a number of factors including the vicinity to business activities and employment is located in a central business district. These older, inner suburbs are also inclined to be more established than the older areas, thus tending to have well established infrastructure which coincides with an increase in the opportunities to participate in social activities, attend cultural and sporting events, and many other attractions that the more central areas of Australian capital cities attract. Each of these reasons anecdotally provides evidence to the observed higher median house price in the inner zones of all capital cities across our sample. However as inner zones are already well established there is potential for faster growth in outer regions. This observation is supported by van Dijk et al. (2011) who find that classes of regions in the Netherlands that contain rural area close to cities exhibit faster growth in house prices compared to the more developed urban areas and remote rural areas. Government revenue is made available through the ABS. It is reported in AUD millions on an annual basis. As rates are noticeably constant over time we can consider if government revenue is to change at any given point this would constitute a change in the value of dwellings. Given that dwellings generally depreciate and we control for the cost of owning and constructing housing stock, this change in value could be attributed to a variation in the value of the space the dwelling occupies. 4 These regions are outlined in greater detail according to the capital city of each state or territory in Appendix XX.

10 Housing stock has been identified as one of the fundamental variables in the determination of house prices in the economic literature. There are a number of ways to account for dwellings stock. Ideally we would wish to examine total stock of housing in the Australian housing market. Unfortunately this variable is not available in the Australian context. We are, however, able to account for the change in housing stock through the variables dwellings completed, dwellings approved, a ratio of the two, and dwellings commenced. Each of these variables are located in the ABS database. There are a number of additional variables identified in the economic literature as fundamentals to house prices. For instance the construction cost of housing accounts for the cost of building and maintaining a property through the producer price index. This supply side effect is accounted for in our empirical model through the producer price index of housing. This captures changes to the construction costs through the appreciation or depreciation in the costs of materials used in the production of housing stock. The following variables have been identified in the literature as factors driving the demand for housing. Population for each region has been calculated from the Regional Population Growth from the ABS. This indicates the number of people in a particular region that require a roof over one s head. We assume that the greater the population, the greater the need for dwellings. Income is accounted for at the state level through the Gross State Product. This indicates that total income available for each state. As we see income rise one would assume that the willingness to pay of individuals would rise, thus increasing the price of houses in a particular state. Real interest rates have been acquired from the World Bank. Interest rates represent the opportunity cost of borrowing money. In a competitive market, where there are low barriers to enter and exit, as interest rates increase the opportunity cost of holding debt raises. It therefore becomes more costly from owners of houses to hold these assets when financed through debt. Hence as interest rates rise, some homeowners may wish to leave the market thus increasing the number of dwellings available in the market. In the following section we will outline the methodology which we use to identify the variables that are of interest in the Australian housing market. First, we control for the interaction between regions in the housing market with the use of fixed effects. Second, we further explore the contemporaneous and lagged effects the variables have on one and other with the use of a panel vector autoregression model. Methodology Drawing upon the literature of fundamentals of house prices we introduce two complementary models to better understand the variables underpinning house prices in Australia. Firstly we model real house prices using regional fixed effects to account for varying regions in Australia capital cities. Each capital city is broken down two three regions (inner, middle and outer), this enables us the use a total of twenty-three 5 cross sections over twelve years which characterizes our longitudinal or panel data. The benefit of using a panel such as this is that is allows for considerably more information to be attained as well as accounting for unobserved individual differences, heterogeneity, that is apparent between city region and across states. We then implement a panel vector autoregression, PVAR, to assess the extent that shocks to 5 The outer region of Darwin has been omitted due to a lack of information.

11 housing variables affect house prices, thus capturing reposes that may not be identified in the first stage of our analysis. In our initial panel analysis we will make use of three different modelling specifications by first controlling for a number of variables clearly identified in the literature before introducing housing stock variables and government revenue variables which account for previously unidentified pressures on house prices that can be attributed to the underlying effect of land on the house price equation. RHP!" = β! + β! HS!" + β! R!" + β! C!" + u!" The first of three models as described above is a pooled OLS model where real house prices (RHP), is regressed on a vectors of housing stock (HS), government taxation revenue (R), a set of control variables (C); u represents the error term of our regression. For the vector HS we consider dwellings completed, dwellings approved and a ration of the two. Government taxation revenue, R, comprises of numerous tax revenue variables of which one is at a local government level in municipality revenue. The additional taxation revenue variables contribute to the real amount of state government revenue generated the sale and/or occupation of dwellings. These include taxation on land, duties on conveyances, other stamp duties, financial transaction revenues, revenue generated through government guarantees and other revenue generated through the transaction of housing stock. Municipality revenue is also included, indicating the amount of revenue generated at a local government level through the housing sector. Finally we include a number of control variables, one of which, producer price index, accounts for the supply side construction costs of building and maintaining properties. There are also a number of demand variables including population of regions, gross state product, real interest rates and the unemployment rate of capital cities. This model makes the usual assumptions of pooled least squares of a zero mean, homoscedasticity, that the errors terms are uncorrelated, and these errors are uncorrelated with the explanatory variables. Due to the panel nature of our data we are required to relax the assumption of zero correlation in the error term over time for the same region. Using this pooled model with the presence of heteroskedasticity and the correlation that we have described above, given the other assumptions are met, will still yield consistent results of the coefficient β yet bias will be present in the standard errors. To avoid this issue we make use of robust standard errors. In the empirical method we make use of both robust and cluster robust standard errors (Hill, RC, Griffiths & Lim 2010). Given that they did not yield significantly different results we have only reported our findings for robust standard errors. Examining the Australian housing market we can observe that each capital city is intrinsically different in the structure of their housing market. These regions are of different sizes, geographical locations, and

12 made up of different ethnic groups, all of which contribute to varying characteristics and requirements of housing dwellings. This can consequentially is observed in dependent variable. Figure 1: Real House Price ($000) heterogeneity across regions The characteristics of each region contained within the panel may, or may not, impact on the independent and consequentially the dependent variables. For the above reasons we have determined that it is appropriate to model house prices in Australian capital cities whilst controlling for cross sectional!"#$%&' heterogeneity using a fixed effect dummy α!. Hence yielding the equation below RHP!" = β! HS!" + β! R!" + β! C!" + α!!"#$%&' + u!"

13 Figure 2: Real House Price ($000) heterogeneity across time Source: Real Estate Institute of Australia Note that cross-sectional ID s 1-8 indicate inner regions, 9-16 are middle regions and are outer regions As observed Figure 2 real house prices are trending upward over time with some movement between periods. This trend may indicate the presence of stationarity and will be investigated presently. On the other hand the movement apparent between periods could suggest time variant characteristics. Since we are estimating our model with real variables one would think that possible inflationary movements between periods had been removed. However to explore this further we will also model time as well as fixed effects shown in model RHP!" = β!!" HS!" + β!!" R!" + β!!" C!" + α!!"#$%&' + γ!!"#$% + u!" Endogeneity and Contemporaneous correlation are no doubt imbedded in the housing sector. This issue is not addressed in panel analysis above. This provides impudence for the use of panel vector autoregression (PVAR) to explore the endogeneity and contemporaneous correlation in the variables of interest. To investigate this further we identify the transmission of housing shocks using a PVAR developed by Love and Zicchino (2006). The benefit of this model is that it accounts for the idiosyncratic nature of regional housing markets by introducing fixed effects (Region! ), isolating the response of real house prices changes to market shock all the time allowing for unobservable regional housing market heterogeneity. This is expressed as

14 y!,! = A L y!" + Region! + ε!", where A L is the lagged operator and y!" is a vector of macroeconomics and housing market variabels. The forward mean is removed using the Helmert procedure following Love and Zicchino (2006) allowing us to avoid obtaining biased coefficients that may arise due to correlation between the fixed effects and regressors. This process allows us to preserve the orthogonality between the transformed variables and the lagged regressors, making it possible to use the lagged regressors as instruments for and estimate Eq. X by system GMM (Arellano and Bover, 1995). Cholesky decomposition is implemented to identify orthogonal shocks in variables of interest and examine their effect on the remaining variables while holding all else constant. We use impulse response functions (IRFs) to analyse the response of one variable to an orthogonal shock of another variable. The confidence intervals for these IRFs are generated using Monte Carlo simulations and identify the response to one shock at a time while holding other shocks constant. Variables which enter Eq. X first are assumed to be most exogenous and therefore affect subsequent variables both contemporaneously and with a lag, while variables that are order later are less exogenous and affect previous variables only with a lag. Discuss the ordering of the PVAR, I believe that I will have to reverse the ordering of the PVAR. Most exogenous first.discuss

15 Results Fixed Effects Model Our results, found in the appendix tables, are reported as the three panel models specified in the previous sections. We firstly use pooled OLS as a benchmark before incorporating regional fixed effects in the second model and additionally time fixed effects in the third model. The dependent variable is the change in real house prices is in $ 000AUD. In Table 1 we regress changes in real house prices on dwellings completed, a vector of government taxation revenue variables and a vector of controls. We can see that model two outperforms the OLS model one once we introduce regional fixed effects allowing for any unobserved heterogeneity that may be apparent between regions. When introducing time fixed effects alongside regional fixed effects in model three, the within R-squared improves again. Further examination of Table 1 demonstrates that changes in real land taxation revenue is positive and statistically significant across each of the three models. This indicates that, as the state government generates a greater amount of income from owner occupants residing within a given state, a proportion of this cost the owners bare is pushed on to house prices in Australian capital cities. Here we can think of land taxation as the cost of occupying a residence in a particular state. As the costs increase, the owner occupant either choose to move from their dwelling to a less expensive location or continue to reside in their current location. We attribute the illiquid nature of housing and the large opportunity cost of relocation to be the reason behind the observed positive correlation between government revenue generated from land taxation. Given these constraints, a proportion of an increase in land taxation revenue will be pushed on to the value of the dwelling and land on which it resides. Also contributing positively to changes in real house prices across our panel is the change in real conveyance revenue. This state level taxation is statistically significant at the one percent level and positively correlated with the dependent variable across each of our three models in the tables above. Real conveyance revenue has a direct relationship with the transaction costs of relocating. As the cost of relocating rises a proportion of this cost is directed the value of house prices in a given state. However, Davidoff and Leigh find that in their study of Sydney post code level house prices conveyance rates have a negative effect on house prices (2013). This could indicate presence of an endogenous relationship which will be further explored in the subsequent section. Change in real municipality revenue demonstrates a negative correlation to real house price. This variables represents the cost owners of property bare in order to poses a dwelling in a particular region. Municipality rates vary region to region though in each instance the charge to owners is used in order to fund the cost of operating various facilities in the area such as waste removal or street sweeping. These results indicate that changes in real municipality revenue has a greater impact on changes to real house prices once regional fixed effects been included.

16 Dwellings completed also continues to provide a positive correlation to real house prices and significant at the one percent level. As dwellings completed is the growth of housing supply in Australia, we can consider the change in dwellings completed as the acceleration of housing stock in Australia. Our results indicate that as housing stock accelerates by one unit a positive change in real house prices is observed of As economists we are understand that as supply increases then prices could fall, contradictory to our findings in this instance. However we must remember that in this instance one building does not commence construction of a new dwelling unless that dwellings has an occupant waiting to fill it. It is often observed that developers do not commence construction until a certain proportion of their development is pre-sold. Thus an increase in supply is always accompanied by an increase in demand. Thus accounting the results observed. Our findings indicate that interest rates have a strong and positive correlation with changes in real house prices. Therefore, decreases in interest rates would have a dampening effect on house prices. One would believe that given houses are an asset and we were to think of them akin to the stock market. If interest rates were to raise this would indicate a rise in the opportunity cost of borrowing money and holding debt. Thus a proportion of the stock market will find it too costly to hold shares, offloading their holding, resulting in a fall in asset prices. When we consider the housing market this is also partially true, a raise in real interest rates will constrain considerable proportion of home owners who finance the ownership of their dwelling through debt. However due to the illiquid nature of houses the market is not able to clear so easily. Not only is it difficult to sell a house quickly there are also the additional opportunity costs of relocating such as finding a new job, new schools for the children of the family, developing new social networks, not to mention the time, money and funds that are required to locate a suitable dwelling. Additionally a significant proportion of the owners mortgage repayment compensates the owner for rent that would otherwise be incurred if their where to live in a property owned by another. Although this positive correlation is contrary to the theoretical explanation it is found to be extremely consistent across each of our models and corroborated in a number of other contemporary studies. Positive correlation between interest rates and house prices has also been demonstrated in a recent study of 12 OCED nations, including Australia (Dilan 2014). From their estimates of panel cointergration relation Dilan et al conclude that real house prices are dependent on real GDP per capita, the real long term interest rates and the global stochastic trends. This finding is farther supported through a consecutive studies one which considers Australia s closest neighbouring nation New Zealand (Shi, Jou & Tripe 2014) as well as the UK (Tse, Rodgers & Niklewski 2014). However there is conflicting evidence in findings that suggest more emerging markets, such as China, exhibit that interest rates operate as a more traditional market clearing mechanism, where mortgage rates have a negative correlation with house prices (Zhang, Hua & Zhao 2012). They find that lower interest rate, faster money supply growth and loosening mortgage down payment requirement tend to accelerate the subsequent home price growth, and vice versa (p. 2360).

17 In order to reinforce our results we regress the variables of interest, as identified above, individually. Table 2 demonstrates that dwellings completed is consistently positive indicating that the rate at which housing supply changes in Australia has a significant baring on changes in house prices. Turning our attention to government taxation revenue in tables 3-5 we can see that each of our government revenue variables of interest remain significant in model 3. It is important to note that changes in municipality revenue only becomes significant once regional and time fixed effects are allowed for. Changes in real land taxation revenue is strongly significant across each model though the value of the coefficient falls in model 3. Changes in real conveyance revenue loses some of its significance in model 3 yet remains consistent across each model. Overall, our model performs considerably better once time and regional fixed effects are included. This indicates that regional influences are important when considering growth of house prices in Australian capital cities. Panel Vector Autoregression Results in the previous section indicate that endogeneity and contemporaneous correlation may be at hand in our fixed effects models. Acknowledging this issue, in this section we explore the endogenous relationship between the variables we have identified in the previous section as having a significant relationship with changes in real house prices, as well lagged changes in real house prices on contemporaneous house prices. We report impulse response function from a panel vector autoregression following Love and Zicchino (2006). The variables of particular interest to us are the lagged shocks to house prices. In figure 1 these are located in the fifth row. Our results indicate that a standard deviation shock to land taxation revenue yields an initially positive shock which turns slightly negative in the first period before a positive spike in the second period, correction and dampening overtime. A shock to real interest rates has an initial positive shock on changes to real house prices though this dampens over time. Likewise changes in lagged real house prices has a positive effect on its contemporaneous variable which dampens over time. Lagged municipality rate revenue also demonstrates a positive shock to changes in real house prices at 1 period before dampening over time. Lagged dwellings completed and conveyance revenue appear to be a little more volatile. Lagged dwellings completed appears to have an initial positive affect on changes in real house prices which turns negative in the second period before stabilising. While lagged changes to real conveyance revenue has a large spike in the first period before a large correction at period two then stabilises over time. Robustness Checks To check the robustness of our results we split our sample into two subsamples. Firstly we repeat our empirical methodology for Melbourne and Sydney and then for regions within all other capital cities except Melbourne and Sydney. We find our results to be most consistent in the non- Melbourne and Sydney sample depicted in table 8. When we incorporate time as well and regional fixed effects in the Melbourne and Sydney sub sable shown in table 7 a number of variable loose significance. This is an area of which further attention is needed. Likewise when conducting our PVAR analysis for our sub

18 samples we find the results of non- Melbourne and Sydney to be consistent with our previous finding while the Melbourne and Sydney sub sample has a very explosive upper bound. Conclusion In this paper we have made considerable contribution to the understanding of house prices in Australian capital cities. By incorporating government revenue into the house price model we are able to account for some of the value that owner occupants attribute to the acquisition and continued residence within their current dwelling. To explore house prices in Australian capital cities we attained a novel data set that decomposed house prices into 23 metropolitan regions of Australia s capital cities. These where made up of an inner, middle and outer region for each. Controlling for the regional and time fixed effects apparent in our data we established a house price model that incorporated a key supply variables, a vector of government revenue and a set of controls as evident in the current literature. Our results indicate that dwellings completed are the best indicator of the rate at which supply of housing stock changes in Australian capital cities. We also found three key government revenue variables that contribute to changes in house prices. Land tax and stamp duty conveyance revenue both have a strong and significant correlation with house prices indicating that a proportion of a positive change to the cost of occupying the land in which a dwelling resides and the opportunity cost of purchasing a property within a given region will be passed on to the value of the property. Municipality rates also contributed significantly to house prices, though in a negative fashion. Panel vector autoregression results explore the endogeneity and contemporaneous correlation issues evident in our static model. These strengthen our original findings and explore the lagged effect of a shock to our key variables of interest. Land taxation revenue shocks and conveyance revenue shocks both have a significant initial effect before dampening over time while. Appendix Tables and Figures

19 Table 1: Change in House Price on All Variables (1) (2) (3) Pooled Regional Effects Regional & Time Effects VARIABLES Δrhp Δrhp Δrhp Housing Stock ΔDwellings Completed ( )*** ( )*** ( )*** Government Revenue ΔReal Land Tax Revenue (0.0246)*** (0.0222)*** (0.0278)*** ΔReal Municipality Rate Revenue (0.0481)** (0.0639)** (0.0793)** ΔReal Other Tax on Housing Revenue (0.106) (0.119) (0.126)** ΔReal Tax on Financial Transaction Revenue (0.0238) (0.0282) (0.0256)** ΔReal Government Guarantee Revenue (0.0550) (0.0686) (0.0787) ΔReal Conveyance Revenue ( )*** ( )*** ( )** ΔReal Other Stamp Duty Revenue (0.0372) (0.0377) (0.0425) Controls ΔRegional Population -8.86e e ( ) ( ) ( ) ΔReal Gross State Product e ( ) ( ) ( ) ΔHousing Producer Price index (1.549) (1.581) (3.325)* Interest Rates (1.775)*** (2.149)*** (3.497)*** ΔUnemployment (Persons) (3.739) (4.702) (4.454) Constant (9.635) (27.29)*** (132.9) Observations R-squared Number of Regions Region FE NO YES YES Year FE NO NO YES Δ identifies a variable in first differences. rhp represents our dependent variable, changes in real house prices. FE stands for fixed effects. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. R-squared is reported as the within R-squared results from our fixed effects model, this is the measurement of correlations squared where y = y!" y! = x!" x! β.

20 Table 2: Dwellings Completed (1) (2) (3) Pooled Regional Effects Regional & Time Effects VARIABLES Δrhp Δrhp Δrhp ΔDwellings Completed ( )*** ( )*** ( )*** Controls ΔRegional Population 3.17e ( ) ( )** ( )** ΔReal Gross State Product ( ) ( ) ( )* ΔHousing Producer Price index (1.496) (1.584) (3.957)** Interest Rates (1.670)*** (1.731)*** (3.680)*** ΔUnemployment (Persons) (3.480) (4.256) (3.761) Constant (8.778) (17.31)*** (125.5) Observations R-squared Number of Regions Region FE NO YES YES Year FE NO NO YES Δ identifies a variable in first differences. rhp represents our dependent variable, changes in real house prices. FE stands for fixed effects. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. R-squared is reported as the within R-squared results from our fixed effects model, this is the measurement of correlations squared where y = y!" y! = x!" x! β.

21 Table 3: Real Land Tax Revenue (1) (2) (3) Pooled Regional FE Regional & Time FE VARIABLES Δrhp Δrhp Δrhp ΔReal Land Tax Revenue (0.0229)*** (0.0258)*** (0.0245)** Housing Supply ΔDwellings Completed ( )*** ( )*** ( )*** Controls ΔRegional Population 6.42e ( ) ( )*** ( )** ΔReal Gross State Product ( ) ( )* ( )* ΔHousing Producer Price index (1.652)* (1.726)* (3.838)** Interest Rates (1.686)*** (1.732)*** (3.655)*** ΔUnemployment (Persons) (3.494) (4.214) (3.727) Constant (9.373) (18.44) (127.2) Observations R-squared Number of Regions Region FE NO YES YES Year FE NO NO YES Δ identifies a variable in first differences. rhp represents our dependent variable, changes in real house prices. FE stands for fixed effects. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. R-squared is reported as the within R-squared results from our fixed effects model, this is the measurement of correlations squared where y = y!" y! = x!" x! β.

22 Table 4 Real Conveyance Tax Revenue (1) (2) (3) Pooled Regional FE Regional & Time FE VARIABLES Δrhp Δrhp Δrhp ΔReal Conveyance Revenue ( )*** ( )*** (0.0107)* Housing Supply ΔDwellings Completed ( )*** ( )*** ( )*** Controls ΔRegional Population -5.85e ( ) ( ) ( ) ΔReal Gross State Product ( ) ( ) ( ) ΔHousing Producer Price index (1.397) (1.448) (3.670)** Interest Rates (1.695)*** (2.175)*** (3.825)*** ΔUnemployment (Persons) (3.545) (4.302) (4.024) Constant (9.143)** (25.91)*** (141.0)** Observations R-squared Number of Regions Region FE NO YES YES Year FE NO NO YES Δ identifies a variable in first differences. rhp represents our dependent variable, changes in real house prices. FE stands for fixed effects. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. R-squared is reported as the within R-squared results from our fixed effects model, this is the measurement of correlations squared where y = y!" y! = x!" x! β.