Analyzing Market Attraction: Focus on the Housing Market

(1)

Analyzing Market Attraction:

Focus on the Housing Market

Master thesis within Economics

Author: John Stenson

Tutor: Mikaela Backman

Pia Nilsson

(2)

Acknowledgments

In order to finalize this thesis I have received support from many sources. I would especially like to thank my two supervisors for giving me plenty of guidance and inspiration. Along with my two seminar colleagues for their useful feedback. Finally, a big general thank you to friends and family for their support and necessary disruptions

during this semester.

Jönköping, Sweden, May 2014 John Stenson

(3)

Master Thesis in Economics

Title: Analyzing Market Attraction: Focus on the Housing Market

Author: John Stenson

Tutors: Mikaela Backman

Pia Nilsson

Date: May 2014

Keywords: Regional Housing Market, Market Attraction, Tobin’s Q, Market Determinants

Abstract

The purpose of this thesis is to analyze the attractiveness of the regional housing market and the factors effecting it. The analysis focuses on municipalities in Sweden and is based on the appliance of Tobin’s Q as a measurement of regional attractiveness, inspired by the previous work of Berg and Berger (2006). Nine market determinants were identified and analyzed with respect to regional attractiveness. Out of these, a well-educated labor force, along with the amount of consumer services available within a municipality and its proximity to the coastline proved to induce the most significant influence. Additionally, some evidence of varying impact from the factors was found across urban-rural range.

(4)

List of Figures

Figure 1: Tobin’s Q Distibution 2000 ...5

Figure 2: Tobin's Q Disbribution 2010 ...5

Figure 3: Trend of Variables included in Tobin's Q ...6

Figure 4: Trend of Tobin's Q by Regional Type ...7

Figure 5: Distribution of Coastal Region ... 13

Figure 6: Diminishing Returns of Poulation Density ... 17

Figure 7: Tobin's Q Trend Line for Jönköping and Comparable Municipalites ... 32

Figure 8: Map of Southern Sweden ... 33

List of Tables

Table 1: Expected effect from Market Determinants ... 15

Table 2: Description of Independent Variables ... 20

Table 3: Descriptive Statistics for Dependent Variables ... 22

Table 4: Descriptive Statisicts for Independent Variables ... 22

Table 5: Bivarite Correlation Matrix for Base Year ... 23

Table 6: Mean Values for Sub-Samples ... 24

Table 7: Regression Models: Nation-Wide ... 26

Table 8: Regression Models: Sub-Samples ... 29

Table 9: National Ranking of relevant Market Determinants ... 34

(6)

1

1. Introduction

Whether to purchase a house or not is often one of the greatest financial decisions a household has to make and numerous factors influencing social well-being must be considered (Straszheim, 1975). Households are challenged to match their financial constraints to preferences of house attributes and location upon making their decision. Since attributes of one single house can involve endless amounts of potential choices and individual preferences, it can prove difficult to analyze on a larger scale. An alternative approach is to analyze the factors affecting the choice of location

concentrating on aggregate measures. The choice for this thesis is therefore to focus on location determinants. The primary target is to enhance the understanding of the Swedish regional housing markets by identifying several location determinants and studying their potential influence on the attractiveness of the housing markets. This subject of study is of great importance as local governments put great emphasis into issues regarding regional housing markets.

The market analyzed in this study is defined as the Swedish market of one-dwelling houses and detached two-dwelling houses. The observed willingness to invest can be theoretically measured by Tobin’s Q and interpreted as a degree of attractiveness to the market. Tobin’s Q is a ratio between the cost of existing house stock and cost of new housing estate (Berg and Berger, 2006). A higher Tobin’s Q value suggests a more attractive market to investors, since the cost of constructing new housing estate is comparatively lower than the price of existing house stock. Simply analyzing the differences in Tobin’s Q between municipalities generates a fair overview of regions’ attractiveness. However, the main focus of the study will be to analyze what different factors affect the attraction level (i.e. the Tobin’s Q-values). To perform such analysis a string of regression tests will be performed.

Thus, the questions raised are how attractive the regional housing market is and above all what factors affect the level of attractiveness. The thesis differ from previous research in the manner it applies the Tobin’s Q to a housing market analysis. While previous research has, for instance, analyzed the housing market in regards to house prices (Johnes and Hyclak, 1994; Quigley, 1999), this thesis will construct an analysis with the observed Tobin’s Q-values as a dependent variable in the regression models, in order to identify the factors that influence regional attractiveness. Furthermore, the regressions will be

(7)

2

analyzed across two regions types in hope of generating a comprehensive understanding of the effects.

Additionally, a case study regarding the municipality of Jönköping will be included. This region, located far from the attractive coastline and Sweden’s larger metropolitan areas, has still experienced a rising housing market ever since the beginning of the 21st century, making it an interesting matter of study. The case study helps confirm the rapid rise on the housing market through a comparison of similar municipalities. Furthermore, the findings suggests that the observed Tobin’s Q of Jönköping is higher than what can be expected from the regression results.

The empirical analysis found evidence that the majority of the identified market determinants influenced regional attractiveness in line with the expectations gathered from theory. Furthermore, some empirical evidence of differences between region types was found after analyzing the municipalities across urban-rural range. Primarily, the analysis has confirmed the important influence of human capital, accessibility to consumer services and the proximity to shore on regional attractiveness. Suggesting local governments may gain from expanding their supply of consumer services and focus on educational institutions.

Initially, the thesis will discuss the theoretical background of the Tobin’s Q and how it is applied to the housing market, followed by a presentation of the identified market determinants and their effect on regions. This will all be included in Section 2, which for convenience will be shortly summarized in section 2.3. The methodology will be presented in section 3, including the method of application for all variables, the empirical design and hypotheses. Section 4 will present the results and analysis from the regression models. And a case study of the municipality of Jönköping will be presented in section 5, followed by a conclusion of the thesis in section 6.

(8)

3

2. Theoretical Background

There are few previous studies focusing solely on factors affecting the Tobin’s Q in the housing market. Most previous research on the housing market is based purely on variables affecting house prices. However, as the price level is incorporated in the theory of Tobin’s Q (Berg and Berger, 2006), several of the factors affecting the supply, demand and consequently the price levels of the housing market, will be implemented in this study.

2.1 Tobin’s Q and the Housing Market

The use of Tobin’s Q is most common when discussing financial assets and can be defined as a ratio of a firm’s market value in relation to the cost of replacing its assets (Chung and Pruitt, 1994). However, ever since the original theory was carried forward by James Tobin in 1969 the technique has been considered transparent and can be applied to several markets, including the housing market (Tobin, 1969). Since Tobin’s Q is first and foremost an investment measurement the majority of previous studies relating it to the housing market prefer to analyze the relationship between Tobin’s Q and housing investment (Jaffe, 1994; Jud and Winkler, 2003; Berg and Berger, 2006). Their results verified the utilization of Tobin’s Q when analyzing the housing market since a positive correlation between increasing tendencies in the Q-values and the willingness to invest in new housing construction was found, indicating that markets showing signs of increasing Q-values is attractive to investors. Another common use of Tobin’s Q is in regional development reports created by public organizations. For instance, in a report by the Regional Development Council in Sörmland (2012) an experimental study was performed on how the Q-value of a region is affected by different traffic network scenarios.

The theory is applied to the housing market by relating the marginal price of the house to the marginal production cost, generating a marginal Q-value. Several authors, namely Hayashi (1982) and Berg and Berger (2006), discuss the implication of a marginal value stating that it is difficult to empirically observe and the only form of observable Q-values is an average Q-value. Hayashi’s (1982) demonstrated that under the assumption that firms are price takers and experience constant return to scale for production and installation the marginal Q-value will equal the average. Berg and Berger (2006) accepted these assumptions and applied the theory to the housing market.

An average Q-value of one suggest that the investment will be repaid without any profit. Any value above one implies that the market is in fact attractive and encourages suppliers

(9)

4

to continue investing in the market, hence the market experience excess demand. On the contrary, an average Q-value below one indicates an excess supply (Berg and Berger, 2006). The long-run equilibrium for the Q-value will be one since in a situation with excess supply (Q<1) the need for new construction will be discarded, which triggers asset prices to increase and eventually return to long-rung equilibrium (Q=1). On the other hand an excess demand (Q>1) will lead to more construction, leading to a decrease in asset price, bringing the market back to long-run equilibrium (Jaffee, 1994). In the report by Regional Development Council in Sörmland (2012) the Q-values are interpreted as measures of performance and attractiveness of a housing market. However, the report suggests that investors in practice should only consider Q-values marginally higher than one in order to expect a profitable investment.

The main advantage of using Tobin’s Q in this kind of analysis is to exploit the convenient interpretation of the ratio. Since the technique is adaptable to different markets this allows the thesis to interpret the Q-value as a measurement of attractiveness towards a regional housing market. As an alternative choice simply implementing actual house prices as a form of regional attraction measure have sufficed in many previous studies. The argument for using house prices is often that the construction cost is integrated in the dynamics of house prices thus suggesting a ratio, such as Tobin’s Q, is needless and the change in existing house prices is enough to recognize the changing market environment (Takala and Tuomala, 1990). However, opting to use Tobin’s Q protects the measurement from misleading results that may occur when ignoring the variance of construction cost across space (Öner, 2014). The construction cost of houses should, like house prices, vary across regions since consumption patterns and economic structure differ between municipalities (Öner, 2014). Furthermore, applying the ratio generates a clear market equilibrium as a point of reference (Q=1), hence simplifying comparisons between regions or regional types. Berg and Berger (2006) claim that one reason for the scarcity of previous studies applying Tobin’s Q to the housing market is due to lack of data. In the case of Sweden, both national and regional data is available to calculate Q-values. The municipal distribution of these Q-values are presented figure 1 and 2 below. In order to establish a general idea of the positioning of the more attractive municipalities and the overall market changes during the past decade.

(10)

5

Figure 1 demonstrates the Q-values for 2000 while Figure 2 represents 2010 values. The

darker colors indicate higher observed Q-values. Approximately 80 percent of the municipalities generated a positive change since the beginning of the 21st century and in particular, as the figures indicate, in the southern part of Sweden. Furthermore, clear tendencies of clustering around the larger metropolitan areas of Stockholm, Göteborg and Malmö can be observed. The municipalities of Jönköping, Norrköping and Helsingborg are also included along with its respective Q-value. These municipalities will later be a part of the thesis case study (section 5) and are therefore included.

Figure 3 below, demonstrates the separate historical trends for the variables used to

calculate the set of Tobin’s Q that will be used in this study1. The three variables are presented in the form of the national average of 289 out of the country’s 290 municipalities2.

1_{Variables in equation:}

Average Q = Average unit cost of existing stock (CE) / Average unit cost of new housing estate (CN)

2_{One current municipality (Knivsta) is disregarded throughout the study due to its late establishment}

(2003).

(11)

6

Figure 3: Trend of Variables included in Tobin’s Q

Source: Data extracted from Berg and Berger (2006), chart assembled by author.

Since the average unit cost of existing houses (CE) is observed underneath the average cost of new housing estate (CN) all through the period the general notion of the Swedish housing market is that it is weak on average, especially from 1990 and onwards, until 2005 when the market improved. Prior to 1990 however the unit cost of existing and new houses were relatively aligned implying a higher average Q-value during this time.

As mentioned the thesis will consider the difference in regional attractiveness amongst different regional types. Two region types will be analyzed separately to demonstrate the difference between municipalities recognized as central and peripheral regions3. Distributing the municipalities in this manner will generate an overview of variations in individual preferences and an indication of which factors are more important in each respective area. An initial view of the differences is presented below in Figure 4 in order to demonstrate the assumption that a housing market’s behavior is different across urban-rural range.

3_{Central municipalities are municipalities with a self-serving central district in a functional region.}

Smaller municipalities closely located to one out of the three metropolitan regions (Stockholm, Göteborg, Malmö) are also considered as central regions. While, peripheral municipalities are the remaining municipalities recognized as rural.

See appendix 8.1 for map.

0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8 2 0 2000 4000 6000 8000 10000 12000 14000 16000 18000 1980 1985 1990 1995 2000 2005 2010 To b in 's Q U n it Cos t (SE K)

Tobin's Q Variables: National Average

(12)

7

Figure 4: Trend of Tobin’s Q by Regional Type

Source: Data extracted from Berg and Berger (2006), chart assembled by author.

The figure describes the separate trends of the two regional types with the national average included as a point of reference. Clear signals of an uneven market can be observed as the peripheral municipalities are considered less attractive across the entire period. As assumed the two sub-samples experience similar trends since they correspond to overall market fluctuations. However, the gap between the two shows increasing tendencies from 1995 and onward suggesting that the importance of the contrasting characteristics between the regional types are increasing.

0,5 0,6 0,7 0,8 0,9 1 1,1 1,2 1980 1985 1990 1995 2000 2005 2010 To b in 's Q

Tobin's Q:

Trend by Regional Type

(13)

8

2.2 Factors Determining Market Attraction

Previous research has discussed several different factors influencing the attraction of a housing market. For structural reasons, Table 1 (section 2.3) will summaries and divide the determinants into positive and negative externalities on the basis of their expected market influence found in previous research.

2.2.1 Population and Congestion

The market size is considered as one of the key variables to determine the attractiveness of a housing market (Rivera-Batiz, 1988; Quigley, 1999). The size of the housing market can be expressed as the population of a particular area and a positive population growth rate leads to increasing demand for houses, suggesting that the area is attractive.

Population density has also been introduced as an indicator of region attractiveness on

the basis that population tend to cluster around popular regions (Rappaport and Sachs, 2003). When discussing the influences of regional population size on markets, authors point to external economies of scale and the potential benefits for firms operating around large business areas (Henderson, 1972; Quigley, 1998). However, limits to population growth has led to theories discussing optimum city sizes. Henderson (1972) explains the hypothetical optimal city size by weighing the benefits of economies of scale for firms against it disadvantages. The benefits are described as the increasing return to scale attainable for firms entering the market. On the contrary, the disadvantages is expressed as the increased commuting time that occurs when cities expand. The additional commuting time increase the resource cost and as the population continues to grow the industry will eventually experience decreasing return to scale (Henderson, 1972).

Brueckner (2000) identified three market failures caused by an over-crowded area. First, as also suggested by Henderson (1972), increased traffic congestion leads to increased commuting time for each additional vehicle. Furthermore, a growing region in size and population will demand a larger and more comprehensive traffic system, inducing large costs (Emmerink et al., 1998). Secondly, the social value of open space and nature is often overlooked in the chase for financial benefits, as constructing an excessive amount of new infrastructure or houses may harm the attractiveness and demand for houses in a region (Nilsson, 2014). In such a case the market failure is identified as market forces causing damage to the region by compromising its open space. The third market failure discussed by Brueckner (2000) is the fact that the additional infrastructure associated with new

(14)

9

house construction (e.g. new roads and sewers) increases governmental costs and leads to increased tax rates. The tax increase affects the entire region, generating an unbalanced market since the new homeowners do not pay the full price for the additional infrastructure but are instead charged more for the house itself. Thus, the new profit-making prices attracts new construction leading to an increasingly congested area (Brueckner, 2000). Subsequently, demand on the regional housing market may decrease if the benefits of population size are outweighed by congestion.

2.2.2. Accessibility to Consumer Services

Glaeser et al. (2001) identified several different factors explaining the degree of attractiveness of cities, the first suggesting that a larger selection of goods and services will enhance a city’s appeal. Additionally, Krugman (1980) was early in embracing the concept of “love for variety” based on the acknowledged Dixit-Stiglitz model, which highlights the importance of variety of good and services. Rivera-Batiz (1988) explains the effects of services to regional development by identifying two types of services; producer services and consumer services, and discusses how the variety of services can improve the attractiveness of regions. Producer services are used by industries in need of certain assistance (e.g. repairs, transportation and legal advice) and a rise in population will increase the specialization of these services (Rivera-Batiz, 1988). Consumer services are services directed towards individual consumers (e.g., restaurant, theaters and barber shops) and they follow a different trend from producer services by increasing utility through greater variety. For both types, a population increase will enhance the markets’ ability to satisfy the wide range of services demanded, leading to increased utility (Rivera-Batiz, 1988). A valid approach for this thesis is to solely focus on consumer services since the aim is to analyze region attractiveness from the viewpoint of households. Furthermore, consumer services are non-tradable, hence only dependent on regional demand which simplifies the data sampling (Rivera-Batiz, 1988).

In light of the Dixit-Stiglitz model a utility function for consumer services helps explain the influence of a diversified service stock, demonstrated below:

𝑍_𝑗 = 𝑁(1−𝜃)𝜃 𝐶_𝑗𝑚

The model highlights the effect on utility gained from consuming local services, 𝑍_𝑗, based on the amount of local services offered, N, in regards to consumer preferences, (1−𝜃)_𝜃 and

(15)

10

household demand. A low value of 𝜃 implies a greater variety between the services (N) which increases utility.

Accessibility to markets, for instance markets for consumer services, is an important

factor of economic progress as discussed in the early studies of Marshall (1920). He discusses the simple fact that even locations blessed with abundant resources cannot profit from these if the market for the finalized good is too distant, and thereby explains how firms’ geographical position reflect their productivity4_{. Furthermore, in a study regarding}

market accessibility and market influence Verburg et al. (2011) find clear patterns showing that the influence on regions is most significant around large cities. Thus, theory underlines the importance of accessibility to certain regions (often larger cities) and its markets, as the potential productivity growth will attract firms. And an increase in firms, connected with population growth, will consequently affect the housing market.

Johansson et al. (2002) presents a comprehensive understanding of relevant accessibility measurements, including two variables expected to enhance the attraction of a region in the eyes of households. The two “attraction variables” considered are accessibility to number of jobs and supply of consumer services. The variables are analyzed on three different accessibility levels; (i) accessibility within the municipality, (ii) within a closely related sub-region (FR-region5) and (iii) external accessibility. Both internal accessibility to jobs and consumer services generated positive influences, validating the inclusion of the variables when estimating region attractiveness.

2.2.3. Human Capital

Human capital is considered a positive element for regions. Schultz (1961) was among

the first to discuss the importance of the difference in “quality of human effort” and how it, if invested in, could improve firm productivity and hence affect the housing market as previously discussed. Lucas (1988) provides a comprehensive production function6 in which both the individual rate of human capital accumulation and the external effect of

human capital are theoretically observed. Initially, by extracting the theory of human

4_{Marshall (1920) discussed three effects of firm clustering causing increased productivity; (i) clustering}

of downstream and upstream firms, (ii) labor pooling and (iii) knowledge spillovers (Marshall, 1920).

5_{Sweden’s 290 municipalities are divided into 81 regions recognized as local labor markets or functional}

regions (FR-regions). A functional region usually consists of 4-5 (usually more around the larger municipalities) closely located municipalities. The distribution is based on the intensiveness of commuting between the municipalities (Johansson et al, 2002).

6_{Lucas (1988) production function}_{: 𝑁(𝑡)𝑐(𝑡) + 𝐾̇(𝑡) = 𝐴𝐾(𝑡)}𝛽_{[𝑢(𝑡)ℎ(𝑡)𝑁(𝑡)]}1−𝛽_ℎ

(16)

11

capital accumulation from the complete production function the time spent on accumulating human capital (e.g. attending school) is expressed as:

ℎ̇ = ℎ(𝑡)𝛿[1 − 𝑢(𝑡)]

Implying that the change in human capital, ℎ̇, is dependent upon the already attained stock of human capital, ℎ(𝑡), and the share of time spent on accumulating new human capital instead of producing goods, [1 − 𝑢(𝑡)]. If all time is dedicated towards accumulating human capital (i.e. 𝑢(𝑡) = 0) the stock of human capital increases at a maximum speed (and vice versa). Secondly, Lucas (1988) include the external effect of human capital in his production function to highlight the influence of an educated society. This effect is expressed as the average human capital stock of a population and an increasing stock is assumed to be beneficial for all the production factors (refer to ℎ𝑎(𝑡)𝛾 in footnote 6).

Building on this theory Mankiw et al. (1995) distinguish the difference between

knowledge and human capital to emphasize the importance of analyzing the accumulation

of human capital. The difference is defined as knowledge being a base of understanding how the world works, expressed in text books or by academics, while human capital is the way the population absorbs that knowledge, for example by attending school. Mankiw (1995) recognize knowledge as a public good that can be easily traded and spread across large areas. Human capital, however tends to have more regional traits since it is partly dependent on the quality of education centers in the region, but more notably due to regional knowledge spillovers. The spillovers are most commonly observed in the form of new innovations and its tendencies to cluster around areas with an extensive knowledge base. This is explained on the basis that skilled people, employed in cutting-edge industries have a greater chance of coming up with new ideas and take advantage of the comparative advantages that comes with innovations (Glaeser 2000; Audretsch 2000). Furthermore, as previously mentioned Lucas (1988) capture the influence of regional knowledge spillovers in his discussion regarding external effects of human capital. Rauch (1993) adopts the approach by using average level of human capital when providing evidence of regional differences in both wages and rental rates, affected by higher average human capital levels. Likewise, Farnham et al. (2011) argue that highly educated people are more likely to act on the housing market, based on the premise that people with higher education are expected to receive higher salaries. Therefore, the thesis expect a positive influence on the housing market from human capital.

(17)

12

Additionally, Black and Henderson (1999a) performed a study on the regional impact of population with a four-year college degree and people with a high-school degree. They found a significant positive relationship between the proportion of college educated people and the city size, implying that the more educated people the larger the city size. Shapiro (2006) discuss the fact that areas with more highly educated people will encourage more diversity in consumer goods and services which, as previously mentioned, have proved to attract people. Moreover, highly educated people are more likely to positively affect their surroundings and in different ways improve the area they live in, for instance through participating in local organizations or political systems (Shapiro, 2006).

2.2.4. Natural Amenities and Climate

Glaeser et al. (2001) discuss location amenities by mentioning the attraction generated from the natural beauty and scenery of cities. Brueckner et al. (1999) presents a comprehensive comparison of differences in natural amenities and historical infrastructure effects between cities in the U.S. and Europe. They found that cities in Europe, with Paris as their prime example, seem to use more governmental resources to maintain the historical infrastructure compared to cities in the U.S. Due to these political incitements central Paris has turned into a very attractive area, crowded with high-income earners. While in many of the American cities the suburbs have proved to be more attractive (Brueckner et al., 1999). Moreover, the influence of proximity to water on cities has been discussed by several authors (Brueckner et al., 1999; Koster and Rouwendal, 2013). They discuss the influence in terms of both enhancing natural beauty but also for convenience purposes as cities often originate along the coastline to decrease transportation cost. Koster and Rouwendal (2013) found evidence of higher house prices, as an indicator of a more attractive market, closer to the waterfront. While, Glaeser et al. (2001) found similar trends when examining the effect of access to water with respect to population growth.

Figure 5 below demonstrates the Q-value distribution in 2000 for Sweden’s 289

municipalities with regards to the identified coastline regions7_{. By analyzing the chart we}

7_{A coastline region is a municipality with direct contact to the seashore. 87 coastline regions were}

(18)

13

can see tendencies of correlation between high Q-values and location near the shore, thus implying an attraction towards living in coastline regions.

Figure 5: Distribution of Coastal Regions

Regional differences in climate have also been analyzed and proven to have a relativity strong effect on location decisions (Black and Henderson, 1999b; Hanson, 2001). Especially, for studies conducted on larger areas with greater regional differences, for example globally (Mellinger et al. 2000) and in the U.S (Roback, 1982). Providing evidence of higher income per capita and population density in areas with a more pleasant climate. However, on a regional basis the importance of climate is harder to prove, which can be logically explained by assuming smaller climate variances. For example, in Mellinger et al. (2000) global study they differentiate regions between six main climate categories, stretching from tropical to polar. A similar study in Sweden would only apply a fraction of these categories. Niedomysl (2008) conducted a questionnaire on the importance of different regional attributes in Sweden, with the climate attribute divided into “southern” and “northern” weather based on the duration of seasons, temperature and precipitation. The findings showed that the majority preferred the weather of their current region, implying either some influence upon deciding where to live or that inhabitants tend to adapt to the regional climate.

Additionally, it has proved difficult to find the most usable way to measure climate differences since climate can be interpreted in several ways. Roback’s (1982) study offers a very broad interpretation of the term by using four different climate variables; (i)

Uppsala Linköping Jönköping Örebro Stockholm Helsingborg Göteborg Umeå 0 0,5 1 1,5 2 2,5 To b in 's Q Municipalites

Tobin's Q Distribution:

Coastal Regions (2000)

No Coastline Coastline

(19)

14

snowfall, (ii) heating degree days8_{, (iii) number of cloudy and (iv) clear days. The study}

suggests that the first three variables have a positive correlation with wage rates, indicating that people living in snowier, colder and cloudier regions demand higher wages to endure the poor climate, aligning with previous mentioned studies.

2.2.5. Crime

Crime rates are expected to affect the attractiveness of regions in a negative way since safety issues are often highly prioritized when deciding on residential location (Wang and Li, 2004). Cullen and Levitt (1999) found strong evidence pointing towards negative influences of crime rates on population size. Their study show that a ten percent increase in crime rates leads to a one percent decrease in population size in U.S cities, and that the population decrease is mainly due to people leaving the region and not a decrease in immigration. This emphasizes the importance of safety since households are willing to change their initial choice of location and go through another major transition if they feel unsafe in their current region. Additionally, Ceccato and Wilhelmsson (2013) discuss the influence of crime rates on Swedish property prices. Their latest research is based on a case study similar to this thesis by analyzing the municipality of Jönköping. The most robust test of their study (using data from 2011) indicated that increasing crime rates had a significant negative effect on property prices. This observation agrees with several previous studies implying crime rates interfere with the housing market in a negative way (Hellman and Naroff, 1979; Gibbons, 2004).

Crime rates have been interpreted in several ways in attempts to prove the expected negative influence. As mentioned, some previous research have provided evidence agreeing with the general conception that crime rates acts as a negative externality. However, the general opinion is that the effect is difficult to statistically prove. For instance, Gibbons (2004) use data on five crime categories in his analysis on the effect on the London property market. Out of these, burglary, initially expected to have the largest negative effect on house prices, proved insignificant. While criminal damage (graffiti, vandalism etc.) also included in the model proved very significant. Gibbons and Machin (2008) argue that some of the complication is due to the fact that crime rates are dependent on other regional factors, for example linked with regional income since it may be assumed that low-income areas experience a higher frequency of crimes. Or increasing

(20)

15

with the rate of business activities in an area (Bowes and Ihlanfeldt, 2001). It is also assumed that the impact of crime rates is difficult for households to realize the effect from, particularly at the time when they decide to purchase or not (Gibbons and Machin, 2008). Generally, for regions to be rejected due to high crime rates the region has established a bad reputation over several years.

2.3 Summary of Effects from Market Determinants

Table 1 summaries the expected effect of the identified market determinants. Indicating

whether an increase in each respective determinant will effect regional attractiveness in a positive or negative way.

Table 1: Expected effect from Market Determinants

Market Determinants Effect References

Population Density Positive Rappaport and Sachs, (2003)

Congestion Negative Henderson (1972); Brueckner

(2000)

Accessibility to Consumer Services

Positive Rivera-Batiz (1988)

Human Capital Positive Lucas (1988);

Rauch (1993)

Natural Amenities and Climate

Positive and Negative Roback (1982);

Koster and Rouwendal (2013)

Crime Negative Ceccato and Wilhelmsson

(21)

16

3. Method

This section will demonstrate the method of application for the variables and regression results.

3.1 Dependent Variable

James Tobin’s classical theoretical structure of Q-values has been decoded onto the housing market in different ways. Initially, Berg and Berger (2006) explains the calculation of the Q-values as marginal price in relation to marginal production cost. However, in terms of extending theory into practice, data of owner-occupied houses in Sweden was gathered and later calculated in the following manner:

𝑇𝑜𝑏𝑖𝑛′𝑠 𝑄𝑎9=

𝑃𝑟𝑖𝑐𝑒 (𝑚2_{) 𝑜𝑓 𝑜𝑤𝑛𝑒𝑟 𝑜𝑐𝑐𝑢𝑝𝑖𝑒𝑑 ℎ𝑜𝑢𝑠𝑒𝑠}

𝑃𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛 𝐶𝑜𝑠𝑡 (𝑚2_{)𝑜𝑓 𝑜𝑤𝑛𝑒𝑟 𝑜𝑐𝑐𝑢𝑝𝑖𝑒𝑑 ℎ𝑜𝑢𝑠𝑒𝑠}

Jud and Winkler (2003) interpreted the theory differently by merely examining the market price levels and the potential arbitrage opportunities consumers may realize when choosing between new or existing houses. They gathered data from American price indexes of existing and new houses to develop the following equation:

𝑇𝑜𝑏𝑖𝑛′_{𝑠 𝑄} 𝑏 =

𝑃𝑟𝑖𝑐𝑒 𝑜𝑓 𝑒𝑥𝑖𝑠𝑡𝑖𝑛𝑔 ℎ𝑜𝑢𝑠𝑒𝑠 𝑃𝑟𝑖𝑐𝑒 𝑜𝑓 𝑛𝑒𝑤 ℎ𝑜𝑢𝑠𝑒𝑠

Hence, if the price of houses offered to consumers on the existing market exceeds prices for new houses, demand for new houses will increase and Tobin’s Q will be higher than one, hence indicate an attractive housing market.

This thesis will however use a third approach also initiated by Berg and Berger (2006). Unlike the previous suggestion using price levels, this approach measures the unit cost of existing and new housing and derives the Q-values as follows:

𝑇𝑜𝑏𝑖𝑛′_{𝑠 𝑄} 𝑐 =

𝑈𝑛𝑖𝑡 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑒𝑥𝑖𝑠𝑡𝑖𝑛𝑔 𝑠𝑡𝑜𝑐𝑘 𝑈𝑛𝑖𝑡 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑛𝑒𝑤 ℎ𝑜𝑢𝑠𝑖𝑛𝑔 𝑒𝑠𝑡𝑎𝑡𝑒

Implying that if cost of building new houses is lower than cost of existing housing stock, the market is attractive (i.e. Tobin’s Q > 1).

9_{Numerator and the denominator are adjusted for net depreciation and production subsidies, respectively}

(22)

17

The three methods demonstrated of measuring Tobin’s Q are only marginally different and all apply to the same theoretical structure discussed in previous sections. Consequently, the reason for applying this particular approach is partly due to availability of data.

3.2 Independent Variables

The first independent variable is the population variable. A region’s attractiveness is dependent on the density of the population as the majority prefer to live in denser areas until a certain point (Henderson, 1972). Therefore, the variable representing population density is followed by a squared version of itself, specifying the congestion variable. This quadratic test helps to identify the eventual diminishing returns from population growth caused by congestion (Gujarati and Porter, 2009). Figure 6 below demonstrates the non-linear relationship between municipal population and its second order term to highlight the eventual diminishing return from regional population growth.

10

Figure 6: Diminishing Returns of Population Density

The inverted U-shaped form of the line suggests that after certain point, population density may become unbearable for a region’s inhabitants (e.g. due to traffic congestion) and the positive effects from population growth will cease.

The most common way to estimate human capital is to identify the proportion of people with a certain academic background. Although, this approach has been adopted in different ways. For instance Barro and Lee (2013) produce a very complete examination by using four different levels of education when analyzing a population sample, the levels being; no formal education, primary, secondary and tertiary education. However, the

10_{The four largest municipalities (Stockholm, Göteborg, Malmö and Uppsala) are discarded to avoid}

biasness and to generate a clearer demonstration.

Po p u lat ion p er Mu n ici p alit y Population2

(23)

18

thesis will adopt a more common and narrow approach by only consider groups with higher education, for instance like previous studies using the share of people with a college degree and/or high-school degree (Black and Henderson, 1999a; Farnham et al., 2011). This thesis will interpret human capital as the share of the labor force with a three-year-bachelor degree or higher.

The accessibility to consumer services will be represented by two separate variables. One denoting the amount of consumer services offered within the municipality. While the second denotes the amount of consumer services accessible within the FR-region of the municipality. Since these services are non-tradable commodities, quantifying the amount of local consumer services available is relatively straightforward if we assume employed labor within the relevant service industries equals the supply (Johansson et al., 2002). The second variable regarding consumer services is included since the intensive commuting across municipalities sharing FR-region will generate some degree of access to the consumer services offered within the entire FR-region (Johansson et al., 2002).

A dummy variable will be included to represent the identified coastal regions in Sweden. The coastal regions are defined as municipalities with at least one building located within 100 meters from the coastline11. This variable will determine whether the proximity to

water positively affect the Q-values. Additionally, two separate climate variables,

temperature and precipitation will be included. The two climate variables are based on the average regional temperature and precipitation over three decades (1961-1990). The thesis optioned to settle for two climate variables due to the relatively similar climate conditions within the country and also due to the comprehensive data available via the Swedish Meteorological and Hydrological Institute (SMHI). However, SMHI measure climate by using weather station spread across the country and since the municipalities differ in geographic appearance some quantities will unfortunately be localized. Additionally, some of the weather stations represents two or more municipalities (ex: the weather station located at Arlanda represents five different municipalities) which may, even though the weather conditions are not assumed to vary significantly, lead to some bias in the regression.

Crime is as discussed in the theory difficult to measure in an efficient way. For a regional

analysis two main approaches are found to be appropriate; crime rates and crime density.

(24)

19

Crime rates is simply expressed as reported crimes in relation to the population size, while crime density is the ratio between reported crimes and region size. Bowes and Ihlanfeldt (2001) argue that crime density is a more appropriate measurement for studies concerning smaller neighborhoods, due to the higher probability of crimes occurring around business activities which may vary across areas. However, for this thesis using crime rates is a valid approach since the analysis is based on larger areas all including some business activity.

3.3 Empirical Design

The thesis will use an ordinary least square (OLS) regression model as demonstrated below, to describe the relationship between the attractiveness of a region expressed as Tobin’s Q-values, and nine market determinants. The data sample concern 289 out of Sweden’s municipalities and is stretching from 2000-2010. Most data is gathered from Swedish Statistics, for opposing cases the data source will be mentioned.

The regression model used is demonstrated below followed by an explanation of the variables in Table 2.

𝑇𝑜𝑏𝑖𝑛′_{𝑠 𝑄}

𝑖,𝑛−𝑡 = 𝛽1+ 𝛽2𝑃𝑜𝑝𝐷𝑒𝑛𝑖𝑡+ 𝛽3𝑃𝑜𝑝𝐷𝑒𝑛𝑖𝑡2 + 𝛽4𝐻𝑢𝑚𝐶𝑎𝑝𝑖𝑡+

𝛽₅𝑀𝑢𝑛𝐴𝑐𝑐_𝑖𝑡+ 𝛽₆𝐸𝑥𝑡𝐴𝑐𝑐_𝑖𝑡+ 𝛽₇𝐷_{𝑐𝑜𝑎𝑠𝑡𝑎𝑙}_𝑖𝑡+ 𝛽₈𝑇𝑒𝑚𝑝_𝑖𝑡+ 𝛽₉ 𝑃𝑒𝑟𝑐𝑖𝑝_𝑖𝑡+ 𝛽₁₀𝐶𝑟𝑖𝑚𝑒_𝑖𝑡+ 𝜀_𝑖𝑡

With i and t representing the considered municipalities and base year 2000, respectively. And n is year 2005 and 2010.

(25)

20

Table 2: Description of Independent Variables

The variables are all expressed as a share of a relevant sample, thus allowing the variables to be kept in in normal percentage form when running the regressions. A series of regression test will be completed in order to generate a comprehensive discussion of the effects of the market determinants and how the observed effect differ across urban-rural range. Initially, a base year is selected (year 2000) from which the independent variables will be gathered. The first regression test will be a nationwide test analyzed solely on data from base year 2000 in. This regression will be followed by a robustness test on a five and ten year time interval, engineered by changing the dependent variable to the observed change in Tobin’s Q within the two intervals (i.e. Δ 2000-05/10). If the coefficients

12_{Consumer service industries}

SIC codes:

SIC (92): 55111-55529, 92310-92792

13_{Data on climate variables are gathered from the Swedish Meteorological and Hydrological Institute}

(SMHI).

14_{Data on crime rates are gathered from The Swedish National Council for Crime Prevention (BRÅ,}

2014).

Variables Functional Form Description

Tobin’s Q (Y) Unit cost of existing stock Unit cost of new housing estate

Dependent variable representing regional attraction measurement

PopDen (X1) Municipal Population

Area (𝑘𝑚2₎

Regional population density

PopDen2 _(X

2)

[Municipal Population

Area (𝑘𝑚2₎ ]

2 _{Population density squared indicates}

the effect of a congested area

HumCap (X3) Educated people

Total employment in muncipality

Share of work force with a 3-year-degree or higher

MunAcc (X4) Employment CS12

Total employment in municiaplity

Share of the municipal work-force employed in consumer service businesses accounts for the amount of consumer services available

ExtAcc (X5) (Empl CS FR − Region − Empl CS Mun)

Total employment in FR − region

Share of work-force employed in consumer service businesses within FR-region

Dcoastal (X6) 1: coastal region

0: non-costal region

Dummy variable for coastal municipalities

Temp (X7) Average temperature 1961-1990 Average temperature measured in

Celsius from 1961-1990.13

Percip (X8) Average precipitation 1961-1990 Average precipitation measured in

millimeters from 1961-1990.

Crime (X9) Reported crimes

Population

(26)

21

change sign or indicate different levels of significance for the models with the observed change as dependent variable, the variable effect cannot be considered robust and fully reliable. These regression are performed to test the first hypothesis (refer to section 3.4). In order to test the second hypothesis (refer to section 3.4) the municipalities has to be divided into groups representing different region types. The region types compared are

central regions and peripheral regions. The central regions amounts to 139 municipalities

while the remaining 150 are considered peripheral municipalities. The test for robustness constructed by changing the dependent variable will be repeated here.

3.4 Hypotheses

In light of the discussed theory two hypothesis can be generated: - Hypothesis 1:

The influence of the identified market attractors (detractors) is positive (negative)

As part of Hypothesis 1 the market determinants will be analyzed in light of expectations developed from theory to evaluate each respective effect. Refer to Table 1 (section 2.3) for a summary of the expected influence. A market attractor is interpreted as a market determinant inducing a positive effect, while a market detractor negatively affect regional attractiveness.

- Hypothesis 2:

The influence of the identified market determinants vary across urban-rural range

As part of Hypothesis 2 the analysis will concentrate on the differences in behavior among the coefficients in order to evaluate the assumed variances across urban-rural range.

(27)

22

4. Results and Analysis

This section will begin with a scrutiny of the variables descriptive statistics and correlation matrix in order to generate an overview of the variable behavior. Followed by a comprehensive analysis of the regression models. The analysis will be conducted in light of the two hypothesis shaped from the discussed theory.

4.1 Descriptive Statistics

An examination of the statistics for the dependent variables for the considered years can be seen in Table 3.

Table 3: Descriptive Statisics for Dependent Variables

Tobin’s Q Minimum Maximum Mean Median

Standard Deviation

2000 0.23 2.41 0.701 0.590 0.349

Δ 2000-05 -0.11 0.67 0.079 0.050 0.120

Δ 2000-10 -0.14 0.93 0.152 0.120 0.1872

The mean value has increased over the period indicating an overall rising market. Since the mean values are marginally higher than the median a slight positively skewed distribution can be detected. The Tobin’s Q rule of thumb (Q<1 = unattractive market) is also worth taking into account, and by comparing it to the mean value of 2000 it suggests that the national average housing market was unattractive at the time.

Table 4 below demonstrates the descriptive statistics for independent variables of base

year 2000.

Table 4: Descriptive Statistics for Independent Variables

Variables (2000)

Minimum Maximum Mean Median Standard Deviation PopDen 0.270 4008.915 124.490 26.530 417.120 PopDen2 _0.070 _16071372.90 _188887.94 _703.65 _1411163.92 HumCap 0.060 0.840 0.1863 1.156 0.111 MunAcc 0.01 0.180 0.046 0.041 0.022 ExtAcc 0 0.011 0.043 0.045 0.017 DCoastal 0 1 0.3010 0.00 0.459 Temp (°C) -1.90 8.40 5.126 5.80 2.134 Precip (mm) 344 1076.70 637.443 605.600 132.458 Crime 0.002 0.205 0.095 0.096 0.033

(28)

23

PopDen and PopDen2_{generates large standard deviations and is positively skewed,}

evidently due to the large variances between the dense metropolitan regions and the sparse rural regions. The mean of DCoastal implies that 30 percent of the municipalities in Sweden

are coastal regions. The remaining variables seem to be fairly normally distributed.

Table 5 presents the bivariate correlation between all the included variables at base year

2000.

Table 5: Bivarite Correlation Matrix for Base Year

Two or more variables experiencing a combined perfect linear relationship will cause multicollinearity and harm the regression model (Gujarati and Porter, 2009). The threateningly high correlation between Pop and congestion PopDen2 is due the fact that the latter is the quadratic form of the first. Both variables will later be mean centered in order to reduce some of the correlation. Mean centering is implemented by subtracting the overall mean of all observations from each individual observation. This method is common when dealing with polynomial regressions. Each correlation value is tested for significance as demonstrated in the table.

Tobin Q Pop Den Pop Den2 Hum Cap Mun Acc Reg Acc

Dcoastal Temp Precip

PopDen 0.608* PopDen2 _0.429* _0.931* HumCap 0.660* 0.311* 0.123* MunAcc 0.237* 0.066 0.070 0.194* ExtAcc 0.346* 0.197* 0.155 0.310* -0.020 DCoastal 0.481* 0.198* 0.106 0.279* 0.116* 0.149* Temp 0.414* 0.167* 0.071 0.202* -0.185* 0.287* 0.175* Precip -0.045 -0.128* -0.108 -0.127* -0.173* 0.066 -0.069 0.189* Crime 0.337* 0.325* 0.243* 0.102 0.124* 0.142* 0.165* 0.263* -0.023 *Pearson Correlation is significant at the 5% level (2-tailed)

(29)

24

Lastly, prior to running the regressions, Table 6 presents the mean values of each variable for the two sub-samples; central and peripheral municipalities15

Table 6: Mean Values for Sub-Samples

Variable Mean Values Central Peripheral

(n) 139 150 Tobin’s Q (2000) 0.8553 0.5583 Tobin’s Q (Δ 2000-05) 0.1253 0.0351 Tobin’s Q (Δ 2000-10) 0.1945 0.1145 PopDen 224.9278 31.430 PopDen2 _389647.5 _2850.6 HumCap 0.2899 0.1468 MunAcc 0.0517 0.0404 ExtAcc 0.0407 0.0446 DCoastal 0.4029 (55*) 0.2067(31*) Temp (°C) 5.146 5.1080 Precip (mm) 623.8 650.0 Crime 0.1073 0.0835

*number of coast regions

The average Tobin’s Q for base year 2000 suggests that central regions are more attractive than peripheral regions as well as the national average (refer to Figure 3, section 2.1). Additionally, the average increase in Tobin’s Q values is significantly higher for the central regions during both periods, suggesting faster growing and more attractive markets within these regions. The mean values of the two population variables (PopDen and PopDen2) are understandably greater for the central regions. The average human

capital (HumCap) is also higher in the central regions agreeing with theory regarding human capital and its tendencies to cluster around well-endowed regions. The mean values for the two consumer services variables (MunAcc and ExtAcc) suggests an inverse relationship with the central regions experiencing higher accessibility within the municipality, while the peripheral regions tend to be more dependent on supply from adjacent regions. This effect is most probably due the distribution of the FR-regions as

(30)

25

most include one larger central municipality (associated with more consumer services) surrounded by smaller peripheral municipalities. Crime show higher values for the central regions presumed to be because the higher population density and greater concentration of businesses in these regions. The mean values for the dummy variable representing coastal regions (Dcoastal) shows the distribution of coastal access between the two region

types, with approximately 40 percent of the central municipalities and 20 percent of the peripheral municipalities are aligned along the coast.

4.2 Regression Models

The two tables in this section presents the coefficients of a total of nine regression tests.

Table 7 includes three regressions on nation-wide data with the first based on data from

base year 2000, followed by two representing the change of each respective time interval.

Table 8 presents three regressions for each sub-groups (total of six regressions) running

the same estimations. To remediate for multicollinearity, the two population variables (PopDen and PopDen2) with observed high correlation have been mean centered. This method lowered their variance inflation factor (VIF). However, as previously discussed the high correlation between the two variables is understandable due their quadratic relationship, hence some multicollinearity is present16. All regressions are tested with Newey-West standard errors (HAC) to minimize the potential threat of both heteroscedasticity and autocorrelation (Gujarati and Porter, 2009). The Newey-West standard errors suffices to this study since the number of observations are large enough and signs of, first and foremost, heteroscedasticity is present17 (Gujarati and Porter, 2009). Heteroscedasticity is a common occurrence when analyzing cross-sectional data. Cross-sectional data concern members of a population at a certain time and these members are often divided into sub-populations (e.g. members of a certain municipality). Since these sub-populations experience varying surroundings the disturbance term in a cross-sectional regression is most likely non-constant, which suggests heteroscedasticity is present (Gujarati and Porter, 2009).

16_{See appendix 8.3 for VIF test.}

(31)

26

Table 7 below presents the coefficient for the initial nation-wide regression tests. The

Newey-West standard errors are displayed in parentheses and each significant coefficient is marked with asterisks. The R2 and F-value are also included in order to highlight the validity of each regression.

Table 7: Regression Models: Nation-Wide

Variable Tobin’s Q 2000 (base year) Tobin’s Q Δ 2000-2005 Tobin’s Q Δ 2000-2010 Constant 0.0105 (0.066) -0.2578** (0.0399) -0.4054** (0.0602) PopDen 0.0006** (0.0002) 2.74E-05 (5.43E-05) -0.0001* (5.81E-05) PopDen2 _-8.28E-08** (3.53E-08) 1.20E-08 (1.23E-08) 1.60E-08 (1.59E-08) HumCap 1.0325** (0.1784) 0.2322** (0.1087) 0.2451** (0.0884) MunAcc 2.6542** (0.5557) 1.2032** (0.3230) 2.6539** (0.4875) ExtAcc 0.7556 (0.5218) 0.3183 (0.2954) -0.3965 (0.6037) Dcoastal 0.1670** (0.0271) 0.0469** (0.0130) 0.1384** (0.0253) Temp 0.0346** (0.0057) 0.0173** (0.0031) 0.0419** (0.0052) Percip 0.0002* (8.04E-05) 8.91E-05* (3.32E-05) 0.0002** (6.50E-05) Crime 0.3484 (0.4175) 0.6585** (0.1596) 0.4719 (0.2944) R2 _0.77 _0.53 _0.54 F 102.74 35.22 37.38

**significant coefficients at the 5% level * significant coefficient at the 10% level (Newey-West Standard Errors in parentheses)

Beginning by analyzing the R2 values, each test indicates a fairly good model fit. Adding more variables may enhance the R2 but brings a risk of increasing forecast errors (Gujarati and Porter, 2009). The high F-values show signs of an overall significance for the regressions since the general hypothesis, stating that together all the independent variables have zero effect on the dependent variable, can be rejected (Gujarati and Porter, 2009).

Initially, analyzing the base year regression confirms that seven out of the total nine explanatory variables indicate significant values. Population density (PopDen) suggests a small positive impact, while congestion (PopDen2_{) suggests a small negative impact}

(32)

27

(Henderson, 1972; Brueckner, 2000). Neither variable manage to stay constant in sign or significance across the robustness test. Hence, conclusions regarding these variables should be made with caution. Human capital (HumCap) remain significant and positive through each tests implying that a higher educated people labor force is likely to positively effect a region, for instance by increasing productivity rates as suggested by Lucas (1988), which subsequently leads to increased demand on the housing market. Additionally, it is believed that isolated influence caused by the attraction of popular academic institutions may increase regions both in terms of population density and human capital, as suggested by the relatively high correlation (0.31) between the two independent variables (refer to Table 5).

Out of the two variables measuring accessibility to consumer services, the internal accessibility (MunAcc) estimates high and positively significant coefficients over all three tests, while the external accessibility (ExtAcc) fail to impact the dependent variable in a significant way. From a statistical perspective the large impact of the internal accessibility to consumer services can be partly explained by assuming that the variable absorb effects originated from other factors, for instance from population density since the consumer services are associated with the amount of people working in the relevant industries. Nevertheless, the significance results suggests that a greater variety of consumer services offered within a region will enhance regional attractiveness and increase demand for houses, pushing the market towards an excess demand for houses (Q>1). This aligns with Rivera-Batiz (1988) theory concerning utility and its tendencies to increase if the widespread demand of consumer preferences can be satisfied. Johansson et al. (2002) also found internal accessibility to consumer services to positively affect their chosen measurement of region attractiveness (using population growth rates). However, likewise to this thesis their variable for external accessibility to consumer services in adjacent regions (using a comparable FR-region approach) failed to show a significant impact.

The interpretation of Dcoastal implies that since the coefficient remain positive and

significant the regions with a coastline tend to have on average higher Q-values. This agrees with previous research such as; Koster and Rouwendal (2013) findings of higher house prices closer to shore, and Glaeser et al. (2001) verifying the appeal of living near the coastline based on findings of higher population growth around these areas. The two climate variables (Temp and Percip) indicate low but significant coefficients for each regression. Previous studies performed on large areas suggest that people prefer warmer

(33)

28

climate with low precipitation (Mellinger et al 2000; Roback, 1982). For this test the temperature variable supports this theory while the effect from an increased precipitation is instead expected to be negative. The positive effect from temperature can be explained by the higher Q-values associated to the southern part of the county (refer to Figure 1 and

2, in section 2.1). The influence can however be questioned since weather conditions in

Sweden tend not to differ excessively as insinuated by the observed intervals in Table 4. Additionally, previous theory suggest no major variance in influence within the country (Niedomysl, 2008). The Crime variable representing regional crime rates contradicts the expected influence of regional attractiveness since it display positive coefficients. One reason for this can be, as briefly suggested in theory, that crime rates are dependent on other regional factors (Bowes and Ihlanfeldt, 2001). For instance as crime tend to group around business intensive areas it is more likely to occur in central regions often associated with higher Q-values. Furthermore, since the crime variable is based on the total amount of crimes and not focused on the severity behind it, the effect caused by an unsafe environment may be outdone by the fact that regions with a larger population will be punished for more crimes (including lenient crimes, e.g. parking tickets). Nonetheless, since only one out three test show significant coefficients the results cannot be fully reliable.

In light of the regression results the first hypothesis stated in section 3.4 can be answered.

Hypothesis 1 can be fully accepted for variables if they remain significant and consistent

in signs through all tests in Table 7. Hence, the hypothesis test cannot be fairly judged for either ExtAcc or Crime as they fail to impact the Q-values significantly in most tests, therefore the hypothesis is rejected for these variables. Even as both population variables (PopDen and PopDen2) demonstrates the suggested signs for the base year test, neither

manage to produce robust coefficients and should, as previously mentioned, be handled with caution. With this in mind the hypothesis can only be partly accepted for both variables. Percip should according to theory be considered a regional detractor leading to a rejection of the hypothesis since the coefficients produce positive signs. For the remaining variables (HumCap, MunAcc, DCoastal and Temp), hypothesis 1 can be fully

accepted since they stay significant and positive for all tests and should therefore be considered regional attractors.

(34)

29

Table 8 below demonstrates a total of six regression tests performed on the two

sub-groups. The same procedures for dealing with potential statistical violations applied to previous regression models have been repeated here.

Table 8: Regression Models: Sub-Samples

Variable Tobin’s Q 2000 (base year) Tobin’s Q Δ2000-05 Tobin’s Q Δ2000-10 Central Regions Peripheral Regions Central Regions Peripheral Regions Central Regions Peripheral Regions Constant 0.1267 (0.1033) 0.3906** (0.1595) -0.2270** (0.0642) -0.2736** (0.0706) -0.3390** (0.0649) -0.4144** (0.1500) PopDen 0.0005** (0.0001) 0.0014** (0.0003) 1.44E-05 (5.97E-05) 0.0003** (0.0001) -0.0001* (5.41E-05) 0.0008** (0.0002) PopDen2 _-7.18E-08* (3.68E-08) -1.45E-05** (3.66E-06) 1.39E-08 (1.41E-08) -2.42E-06** (1.18E-06) 1.79E-08 (1.55E-08) -6.73E-06** (2.58E-06) HumCap 0.8537** (0.2240) 0.7877** (0.1920) 0.1231** (0.0546) 0.4512** (0.1264) 0.2080** (0.0918) 0.5046** (0.2579) MunAcc 2.7229** (0.7033) 2.1406** (0.6835) 1.1697** (0.6148) 0.6274** (0.2769) 2.3074** (0.7080) 2.3105** (0.6105) ExtAcc 2.4196** (0.8778) 0.5933 (1.5874) 0.1934 (0.3736) 2.2789** (0.7138) -0.9139 (0.8036) 3.0931* (1.6157) Dcoastal 0.1599** (0.0279) 0.1574** (0.0425) 0.0485** (0.0166) 0.0198** (0.0096) 0.1015** (0.0231) 0.1475** (0.0368) Temp 0.0305** (0.0092) 0.0210** (0.0068) 0.0227** (0.0048) 0.0144** (0.0031) 0.0437** (0.0076) 0.0375** (0.0075) Percip 9.40E-05 (9.25E-05) 0.0002* (0.0001) 5.72E-05 (4.69E-05) 0.0001** (4.57E-05) 0.0001* (7.65E-05) 0.0003** (8.34E-05 Crime -0.1571 (0.7742) -0.7552 (0.4699) 0.7252** (0.2890) 0.0011 (0.1893) 0.5450 (0.5193) -0.2630 (0.3682) R2 _0.80 _0.60 _0.55 _0.45 _0.54 _0.63 F 55.57 23.97 17.56 12.48 16.84 26.84

**significant coefficients at the 5% level * significant coefficient at the 10% level (Newey-West Standard Errors in parentheses)

Initially, in regards to the two test statistics included, the R2 shows a similar model fit for the central regions as for the nation-wide models for base year 2000. Even though R2 values for the peripheral regions indicates a worse fit the levels are still acceptable. All the F-values show an overall significance.

Similarly to the nation-wide tests, population density (PopDen) and congestion (PopDen2) show significant and expected signs for base year 2000 for the two region types. Nevertheless still very small, they indicate a diminishing return of population growth as previously discussed. As for the difference between the region types, although

Analyzing Market Attraction: Focus on the Housing Market