Gas extraction in the Groningen field and its effect on the property value:
the design and validation of an improved appraisal method
Abstract
Since the first earthquake caused by gas extraction occurred in the Groningen gas field a broad discussion has grown on the influence of these occurrences on property value. Koster and van Ommeren (2015) show that there is, on average, a significant negative influence on the property value. In this research the outcomes of their research will be discussed and transformed from a global perspective towards an individual property level. This makes it possible to incorporate their findings in the appraisal method for the property. Using the regulative cycles of Van Strien (1997) and BPMN an improved method is implied. This improved design is validated using all the NVM transactions of 2014. By doing this, the research explores the opportunities for hedonic pricing models in monitoring systems. The results from the validation show that the model proposed in this research is partly able to calculate the property value in the Province of Groningen. Furthermore, the model shows that there is a significant negative influence of earthquakes on the property value in the Province of Groningen. This means that the model proposed in this research can be validated following the regulative cycle.
Keywords: Natural gas extraction, Business Process Modelling Notation, earthquakes, design science, hedonic price analysis, property value, appraisal method.
Sam Peetsold S1989286
Sampeetsold@gmail.com / S.peetsold@student.rug.nl
Programme: Technology and Operations Management University of Groningen
22 June 2015
2
Contents
Preface ... 3 Introduction ... 4 Research Design ... 8 Design solution ...9 Implementation ...9 Validation ...10 Diagnosis/Analysis ... 13 Current method ...13Proposed model of Bakker ...13
Identifying improvement points ...14
Design Solution ... 16
Verifying the model ... 18
Property data ...21
Geological data ...22
Test Results ...24
From m2 to total property value ...28
Conclusion ... 30
References ... 31
Appendix 1 Stakeholders, Goals and CSF’s ... 34
Appendix 2: Current appraisal process... 35
Appendix 3: Proposed appraisal method by Bakker (2015) ... 36
Appendix 4: Characteristics and coefficients (Koster & Van Ommeren, 2015) ... 37
3
Preface
4
Introduction
The discovery of the Groningen gas field in 1959, and the start of exploitation in 1963 by the Nederlandse Aardolie Maatschappij (NAM), brought great prosperity to the Dutch state. At the moment of writing, the Dutch state earned a total of more than 200 billion euro. However, this great prosperity came with it side-effects. Due to the extraction of gas from the fields earthquakes were induced. The first earthquake caused by the extraction of natural gas occurred around Assen on 26 December 1986; since then both the intensity and magnitude have increased drastically (KNMI, 2013). In recent years the extent of the earthquakes has intensified and the damage has increased. At the time of writing, the Dutch government is heavily discussing how to cope with the increasing damage from the earthquakes. The conflicting interests of the stakeholders cause the main problem. Whereas the NAM and the Ministry of Economic affairs are unwilling to lower the outflow of gas from the Groningen field, and by doing this increasing the quantity and magnitude of the earthquakes, the Province, municipalities and property owners demand a decrease in the outflow of gas and a better system for the compensation of the decreased value of their property (Bakker, 2015).
Since the whole debate started, the property owners and their advocacy groups gained more recognition for the problem. Progressively the problem is recognised and addressed by politicians. They admit that the current situation has to be changed and that big steps have to be taken to resolve the problems. Recently, the Council of State (Raad van State) ruled that the extraction of gas in Loppersum is prohibited for an indefinite period. Loppersum is the municipality where the strongest earthquake to date has occurred with a magnitude of 3.6 on the scale of Richter (NRC, 2015). Furthermore, the Labour Party (PvdA) has announced that they will pursue a change of the Mining Act. This will reverse the burden of proof so that the NAM has to show that the damage of properties is not caused by their mining operations (Volkskrant, 2015). Even when the extraction of gas stops, earthquakes will still remain in the area due to the “law of preservation of earthquakes”. Hagoort (2015) concludes in his research that earthquakes that have already occurred are just a beginning. He expected that at least 1100 earthquakes with a magnitude higher than 1 on the scale of Richter will occur.
5 located in the eight so-called problem municipalities are eligible1. At the moment of writing only 334 properties were appraised and only 78 properties owners were compensated (NAM, 2015)
For the compensation programme, the Ministry of Economic Affairs and the NAM base their viewpoint on the findings of Ortec Finance, a financial firm, which conducted the research on behalf of the same ministry. The quarterly reports only suggest that in the third quarter of 2013 a significant decline of property value occurred. However, the properties owners, municipalities and the Province disagree with the results. De Kam & Elhorst (2014) identify four limitations of the research of Ortec Finance.
The first limitation is that they compare problem municipalities (i.e. exposed to severe earthquake damage) with connecting municipalities. Because of this, their research is based on a limited amount of observations. Secondly, a number of factors are not included in the hedonic pricing model they use. For example, they do not take into account the effort of earthquake risk reduction measures (ERRMs), which have a positive effect on the property value (Willis & Asgary, 1997). Furthermore, the data used by Ortec Finance are provided by the NVM (Nederlandse Vereniging van Makelaars). The main limitation of this is that the database only covers 70% of all the property sales in the Netherlands. Lastly, the database is not publicly available, which makes it impossible to repeat the calculations. Due the aforementioned conflicting interests of the stakeholders, the University of Groningen started an objective research, which this thesis is part of, to come up with an independent valuation model, which can be used to compensate the affected stakeholders.
Recently, Koster and van Ommeren (2015) were the first to publish a scientific study on the influence of earthquakes on the property value focused on the Groningen field. In their research a hedonic pricing model is used to calculate the influences. They found a significant decrease in property value for properties located in areas where strong earthquakes occurred. Following their research, on average, a property will decrease 1.23% in value for every noticeable earthquake that occurs in the area. This research will use their findings to further analyse the influences and transform the global conclusions to individual cases.
1 The municipalites of Appingedam, Bedum, Ten Boer, Delfzijl, Loppersum, Slochteren,Winsum and
6 Besides the of Koster and van Ommeren (2015), this project builds on the research of Bakker (2015). In his research he identifies the stakeholders, their goals, and the critical success factor (CSF) for each goal. Furthermore, he identifies the current way of calculating the property value and the compensation. The CSF’s and goals are validated through the use of interviews and based on those results he proposed an improved business process model and notation (BPMN) for the calculation of property value in the Groningen gas field and the compensation programme that is currently used by the NAM and the Ministry of Economic Affairs. As can be seen in Appendix 1 all the stakeholders share the following goals: the appraisal method should output usable and reliable data, requires a low level maintenance, and is flexible to changing context.
However, the stakeholders disagree on the amount of depreciation on the property value. In order to improve the current situation, a more objective system has to be established that is able to monitor the current situation and which is adaptable to all the properties in the Province of Groningen. To analyse the problem, and for the establishment of an improved design, the regulative cycle of Van Strien (1997) is used. This five steps methodology will be elaborated more in-depth in the research design part. The current problem results in the following question:
How to construct an appraisal method that captures the depreciation in property value caused by induced earthquakes?
In order to answer this research question the findings of Koster and van Ommeren (2015) will be used. Their research provides the practical input for the appraisal method. In order to design the best workable appraisal method, the current way of appraising a property will be discussed, and by using their hedonic pricing model and Bakker’s (2015) findings, an improved appraisal method will be put forward. For the improved appraisal method, which should make it possible to measure property value decrease without selling a property, there are four stakeholders. First of all, the owner of the property, secondly, the appraiser, thirdly, the NAM and lastly the municipality. First the current BPMN model and the proposed BPMN model of Bakker (2015) are analysed and discussed. From here on the current appraisal method will be analysed.
7 2014 to see whether the findings of Koster and van Ommeren (2015) can be transferred from a global outcome towards an individual outcome. For this research, data from the NVM will be used. The NVM is the Dutch realtors organisation. They collect all the sales that their members make. This is about 70% of all the transactions in the Dutch real estate market. However, there is only one other database, Kadastar, which is less specific about the property characteristics and is not up-to-date.
The main contribution to the academic literature is in the field of the valuation of property depreciation caused by human induced earthquakes. The Groningen field is the first known area in the world where there are signs of a significant influence on property value due to induced earthquakes. Because of this, there is a shortage of literature for parties to base their appraisal methods on. This intensifies the problem of subjectivity. The methods and results of this research can be used to calculate the influence of earthquakes on the property value. Part of the research will focus on identifying more characteristics that could improve the model. Furthermore, the analyses of the hedonic pricing model will provide literature focused to what extend a hedonic pricing model based on historical data can be used to estimate the value in the coming years.
8
Research Design
This research fits within the field of design science. In design science there are two types of problems. A knowledge problem which can be defined as “a difference between what stakeholders know about the world, and what they would like to know.” (Wieringa, 2007), and a practical problem, “The difference between the way stakeholders experience the world and the way they would like to experience it”. (Wieringa, 2007). Furthermore, the answer to a practical problem will be characterized as useful or useless or everything in between.
The practical problem of this research is the method that is used to valuate properties that are harmed by earthquakes due to gas extraction in the Groningen field. Currently, properties sold after 25-01-2013 are only eligible for this earthquake compensation programme. Besides this, only eight municipalities are eligible. Most of the property owners are not satisfied with the method. So the practical problem that will be investigated is how the appraisal method can be redesigned in such a way that the property owners do get compensated for the full extent of the damage. The starting point of this new appraisal method is the BPMN model of Bakker (2015). The BPMN model of Bakker is focused on the current method used by the NAM and the realtor to calculate whether there is a difference between the values of properties in the risk areas and reference areas.
In design methods the regulative cycle of Van Strien (1997) is a commonly used structure to perform research. This cycle is focused towards the improvement of individual problem-situations with the help of low-level, problem-directed theories. The cycle consists of five steps, which are shown in figure 2.
9 Design problem
The cycle starts with the design problem. This step focuses on identifying the stakeholders, their goals and the critical success factor (CSF) of each goal. Stakeholders are the parties that can be affected by the design problem. Their goals are the desired changes in the current state of the world and the CSF’s are the success factors that have to be satisfied in order to attain the CSF original goal (Balsters, 2015). In the research of Bakker (2015) the stakeholders of the whole compensation process, their goals and CSF are identified. An overview of these findings can be found in appendix 1. In the next part the five steps of the regulative cycle will be explained.
Diagnosis/analysis
The second step is the diagnosis/analysis. To perform this, two questions have to be answered:
1. What are possible causes of the difficulty in resolving a CSF? 2. In what order does the CSF need to be treated?
Design solution
Based on outcomes of those questions a BPMN model is made. This is part of step three, the design solution. In this step the following three questions are most important:
1. Which solution alternatives are available?
2. Is it possible to gather old solutions to build a new one? 3. Is it necessary to invent a new solution?
Implementation
10 Validation
The last step is the validation of the model. As mentioned in the implementation, the proposed design can be tested using multiple options, which makes it possible to validate without the implementation phase. In this research, implementation is not feasible due to the scale of the project. Besides this, it is important that the valuation model is as accurate as possible, because there is no room for trial and error due to the sensitivity of the situation. To overcome this problem and still validate the design, a model will be designed and built that is able to validate the designed appraisal method. The NVM database provided the opportunity to construct a model that predicts the transaction values for 2014. This model can be used, because Koster and van Ommeren (2015) did not use the transaction data of 2014. This mean that the model can predict the expected value per m2 for the properties sold in 2014, which then can be compared with the historical transaction value. Using the NVM Database, which includes the transaction prices of 2014, the model than can be used to see to which extend it is able to predict the right property value in the Province of Groningen. Furthermore, The Royal Netherlands Meteorological Institute (KNMI) database provides the necessary information, such as magnitude, location and date, on the every earthquake since 1991.
In the next four chapters the steps of the regulative cycles will be performed, first the design problem will be explained, this will be followed by a profound analyses of the current appraisal method, the method proposed by Bakker (2015) and this together will compose the fundament of the design solution that is put forward in this research. This design solution is then validated using the transactions of 2014.
Design Problem
11 In this research the outcomes of the paper of Koster and van Ommeren (2015) on the hedonic pricing model of the influence of earthquakes on property prices are used. This type of model is used to create an accurate predictive model. Besides this, it is the same type of model used by Ortec Finance (2013). In a hedonic pricing model the effects of characteristics on the overall transaction price are measured (Monson, 2009). Hedonic models are typically estimated as single stage equations (Sirmans, Macpherson & Zietz, 2005). In most of the cases the model is structured in a semi log form, in which the price is specified in natural logs and regressed against independent variables. Examples of such characteristics are square meters, year of construction, etc. The advantage of using semi log specifications is that it reduces the problem of heteroskedasticity (i.e. the error term is correlated with the variable) and the coefficients can be easily interpreted as the percentage change in the price given a change in the characteristic (Malpezzi, Ozanne & Thibodeau, 1980).
There are multiple reasons why the research of Koster and van Ommeren (2015) is used as the starting point of this research. First of all, in their article they provide detailed information on how they constructed the model and what all the coefficients of the attributes are. Secondly, they perform multiple tests to check for the robustness of their model. They use different fixed effects, e.g. postal codes, time periods, and construction year, to correct for time-invariant on spatial-invariant variables. They check the occurrence for earthquake on randomness and show that earthquakes are significantly concentrated, which means that they do not occur at random around the province. Besides this they do not use a focus on reference versus risk area as Ortec Finance did. At the time of writing the NAM and the Ministry of Economic Affairs have pointed out eight problem areas. This selection was made on the basis of the magnitude of the earthquakes in every municipality. The reference areas are the areas that are directly connected to the risk areas. However, two municipalities (Groningen & Haren) were excluded because of their different social-economical characteristics. This reduces the validity the research of Ortec Finance. Koster and van Ommeren (2015) use postcode area fixed effects to cope with the limitation of the research of the NAM.
12 the model that includes those regional effects is only 60 basis points higher. Besides the compensation for the depreciation of the property towards the homeowner, a second stakeholder is highly influenced by a lower property value. The OZB (onroerendezaakbelasting) generates about 8% of the total income of Dutch municipalities. Based on the value of a property a certain tax rate provides the tax expenses for the homeowner. Table 1 shows the total income for the eight municipalities that perceived the most severe earthquakes.
Income from OZB taxes 20132
Decrease property value per earthquake (Koster & Van Ommen, 2015)
€ 30,673,159 -2.19%
Table 1: Income from OZB municipalities and influence of earthquakes on property value
Based on the article of Koster and van Ommeren (2015) the average damage of every earthquake for the municipalities can be measured. At the time of writing these municipalities do not receive a special compensation for this decrease in income from OZB. Therefore, the municipalities can be seen as a stakeholder of the appraisal process. For the stakeholders NAM, appraisers and the property owners the CSFs and goals validated by Bakker (2015) are sufficient in-depth because it was already part of the compensation system. For the municipalities it should be clear how they could be compensated for a lower OZB income.
In conclusion, the research of Koster and van Ommeren (2015) has shown that the property value significantly decreased in the Groningen field due to earthquakes. This provides contradicting results with the analysis of Ortec Finance. The model should be more objective and applicable on all the properties in the province of Groningen and not only limited to the eight municipalities. Next to this it should be possible to appraise the damage without selling the property.
2 Based on the year report 2013 of the eight problem municipalities: Appingedam, Bedum, Ten Boer, Delfzijl,
13
Diagnosis/Analysis
The stakeholders their CSFs and goals have to be established and validated. In addition to this, the most appropriate method of valuation has to be selected. On one hand it has to be fast and without a lot of work for the property owners, on the other hand it has to be as accurate as possible. As the research of Koster and van Ommeren (2015) implied a large group of factors affect the property value. Appendix 4 shows the characteristics and their corresponding coefficients. This chapter will focus on the diagnosis of the current appraisal method and its main shortcomings. This will then be used in the next chapter to establish an improved BPMN model to calculate the property value and how much of this value is depreciated due to the earthquakes in such a way that the outcome fits the CSFs and the goals of the stakeholders better.
Current method
In the current method the following steps form the appraisal process:
1. Determine property specific features
2. Check international valuation standard (IVS) of market value of property 3. Collect data value trend in property area
4. Determine 100 reference areas
5. Collect data of value trend in reference area 6. Determine other value determine factors
7. Compare trend in property area and reference area 8. Compare outcomes with Calcasa and Ortec finance 9. Is there a difference in property value?
a. Yes, percentage times selling price is advice b. No, no loss in value is advice
Steps 4,5 and 6 are performed by the NAM and the others by the realtor. The current BPMN of this process is attached in Appendix 2. This BPMN is based on the findings of Bakker (2015).
Proposed model of Bakker
14 not sold it is due to the fact that the market is not willing to meet the price that the owner wants. Literature shows a clear relationship between the listing price, the speed of sale and the magnitude of ultimate transaction price (Knight, 2002). The listing price is the price that is first set when a property is put on the market and affects the transaction price and the amount of time the property remains on the market. Because the data used in this research is based on the NVM, the chance of a listing price far above the estimate value could be seen as rather small. A realtor will only use a listing price for which he could sell the property, before investing in the marketing and effort to sell the property.
Knight (2002) shows that one of the most import determinants of a price lowering is the total length of time the property is on the market. The relationship between the transaction price of a property and the time on the market can be complicated. First of all sellers want to maximize their transaction price while buyers seek to minimize the price they pay (Sirmans, MacDonald & Macpherson, 2010). Because of those two contradicting interests the seller has to find the optimum between the transaction price and the time on the market.
There is a second dilemma for the people in the problem area; the longer a property stays on the market, the more prospective buyers will interpret this as a signal that the property has a non-trivial problem, which in the end will result in a lower transaction price (Taylor, 1991). The current news coverage of the problems on the property market will only increase this effect. Property owners in the problem area are often convinced that they are not be able to sell their property for a reasonable price (Bakker, 2015). This feeling is thus in line with the literature. Besides including the sold/unsold ratio that Bakker (2015) proposed he highlights the role of the municipality. However, he neglects the OZB compensation, which is, as shown before, an important source of income for the municipalities. Even in the calculation made in the design problem there were only eight municipalities taken into account. To solve this difference the municipality has to be included into the depreciation of the properties.
Identifying improvement points
15 However, the research of Koster and van Ommeren (2015) does not show a significant longer transaction time of the property in the areas that have faced strong earthquakes. Furthermore, they did not found a significant difference in the transaction/asking price ratio between affected properties and the properties that were not exposed to strong earthquakes. Their research does not provide any support to include the sold/unsold ratio.
Secondly, the current compensation programme of the NAM and the improvements proposed by Bakker (2015) does not reduce the role of the NAM. The NAM provides a large part of the information for the valuation. The possibility for the NAM to select reference areas gives them the power to act in their own interest. To increase the objectivity of the method the role of the NAM has to be minimized. Another issue that was identified by Bakker (2015) it the requirement for the appraisal method that the property has to be sold before 25-01-2013. The improved method should be able to forecast the property value and above all the depreciation due to the number of earthquakes that a property was exposed to. An appraisal that can monitor the situation in the Province of Groningen will provide better forecasts of the total amount of money that is needed to compensate the property owners. Besides this, it will help to monitor the situation and it will provide the owners of the property with usable data. The reduction in uncertainty for the property owners can help to balance the market again and take away the uncertainty of new buyers.
Lastly, the number of earthquakes a property is exposed to does have an influence. It is important to incorporate the total number of strong earthquakes a property is exposed to (Koster & van Ommeren, 2015). The strength of an earthquake is often measured on the scale of Richter, which is a logarithmic scale that uses the strength of the earthquake. This scale measures the energy of an earthquake. In the research of Koster and van Ommeren (2015) they use the peak ground velocity. The advantage of the peak ground velocity (PGV) is that it measures the intensity of the earthquake on a certain point on the map. This is more applicable for this type of research than the scale of Richter because the distance between the hypocentre and the property do affect the damage. However, there is critics on the PGV. Due to the scope of this research this is not further investigated, however, future research should focus on the best parameter for the earthquake damage. Due to the fact that this research will use the paper of Koster and van Ommeren (2015) the PGV will be used for calculating the exposure of properties to earthquakes.
16
Design Solution
As mentioned in the diagnosis/analysis chapter the research of Koster and van Ommeren (2015) provides a new scientific basis for an improved appraisal method. Based on those findings a new appraisal method is designed. In their research they establish a hedonic pricing model to identify the influence of earthquakes on the property value. The coefficients for the characteristics can be transformed into an appraisal method. To improve the current appraisal method and the method recommended by Bakker (2015) the following model is proposed (Appendix 5).
1) The property owner can request an appraisal.
Currently this is only possible after the property has been sold. This makes compensation impossible for owners that are unable to sell their property. Besides this, an appraisal method that can be used without selling the property will improve the information for the owner and municipality. As mentioned in the diagnosis, currently the municipalities are not compensated for lower income from the OZB due to lower property value.
2) The appraiser then plans a meeting
This meeting is necessary to evaluate the property on multiple characteristics; these are provided by the research of Koster and van Ommeren (2015). The characteristics are shown in appendix 4. Those characteristics all have a significant influence on the property value. Based on the value that is provided by hedonic pricing model a price per M2 is calculated.
3) Identify the number of Earthquakes (PGV >0.5 cm/s)
17 4) Influence of earthquakes on property value
From here on the influence of the earthquakes can be calculated. As given in the research of Koster and van Ommeren (2015) a noticeable earthquake (PGV>0.5cm/s) decreases the value of the property with 2.193%. As mentioned in step 3, the regression will be used without the economic effects in this research. Using the formulas explained in the geological data part the number of earthquakes can be calculated. Only the earthquakes that occurred since 1991 will be measured. After identifying the number of noticeable earthquakes in the postcode area the influence of earthquakes can be measured.
5) Transform findings into whether or not there is a case of depreciation
The outcomes of the appraisal method should be send to the property owner, the NAM and the municipality. The first two are clear stakeholders of this outcome. Based on the outcome a certain compensation system has to be established. How this compensation should be designed will be discussed further on in this research. The municipality uses the property value to calculate the OZB. This provides them with 8% of their yearly budget. Reduced tax incomes for the municipalities have to be compensated.
6) Possibility to dismiss findings
When either the property owner or the NAM is not satisfied, they can protest and ask for a new appraisal. It could be possible that one of the stakeholders does not accept the appraisal. In this situation it is up to an objective party to see whether the claim is legitimate. When this is the case, the findings should be binding for a certain time period. If there are any shortcomings a new process should be started with a new appraiser to increase objectivity.
7) Arrangement of the claim.
18
Verifying the model
With the design solution, the global outcome of Koster and van Ommeren (2015) is transformed to an individual level, i.e., for a property. With the characteristics identified by them and the corresponding coefficients it is now possible to test the outcome for individual projects. As mentioned in the design solution, the valuation is now based on the characteristics and coefficients of Koster and van Ommeren. In their research method and their analyses they base their outcomes on the property market in the Groningen field between 1996 and 2013. They find that on average a property depreciates in value for every earthquake it is exposed to.
Their hedonic pricing model provides the possibility to forecast the transaction price with the proposed appraisal method on the properties that were sold in 2014 to see to which extent their findings hold at the individual level. To do this the same data source as Koster and van Ommeren (2015) is used. In their research they use the NVM property database up to and including 2013. In the introduction it was mentioned that one of the shortcomings of the research of Ortect Finance was the source they used. They also use the NVM database for their models. The downside of this is that de NVM database only covers 70% of the transactions. The Kadaster is the only database that achieves a cover rate of 100%, however this database is far less specific, it contains less characteristics of each property, and moreover it is not as up-to-date as the NVM database.
For a hedonic pricing model a large range of variables is needed, those characteristics are only present in the NVM database. For this reason also this research will use the NVM database. For the valuation the data of 2014 is used. Based on this data, properties will be evaluated on the characteristics and using earthquake information of 2014 from the KNMI. From here on the total number of earthquakes a property is exposed to since 1991 will be calculated. With this information the expected depreciation due to the earthquakes can then be forecasted. With this prediction the historical transaction price will be compared with the forecasted property value.
19 To calculate the difference between those two prices the following steps will be taken. First the data on the property transactions will be analysed. The characteristics that are necessary to calculate the expected value are subtracted from the NVM database. Those characteristics can be found in appendix 4. To analyse the influence of the earthquake the areas that were exposed to earthquakes with a PGV>0.5cm/s have to be determined. For this the database of seismographic activity of the KNMI is used. With all this information the following prediction formula (1.1) is used to calculate the expected value following the model of Koster and van Ommeren (2015):
𝐿𝑁 𝑝𝑖𝑡 = 8.273 − 0.022𝑒𝑖𝑡− 0.460𝑠𝑖𝑡+ 0.023𝑟𝑖𝑡+ (0.0895𝑡𝑖+ 0.148𝑡𝑖+ 0.356𝑡𝑖) + 0.109𝑔𝑖𝑡+
0.122𝑐𝑖𝑡+ 0.1𝑙𝑖𝑡 + (−0.0192𝑏𝑖−0.022𝑏𝑖+0.030𝑏𝑖+0.114𝑏𝑖+0.206𝑏𝑖+0.269𝑏𝑖) + 0.689 (1.1)
For which natural log 𝑝𝑖𝑡 is equal to the property price per square meter. The research of Koster and van Ommeren (2015) provides the constant of the model (8.273). 𝑒𝑖𝑡 is the number of earthquakes that occurred with a peak ground velocity greater than 0.5cm/s since 1991 for a property. 𝑠𝑖𝑡 is the natural log size of the property at 2014. 𝑟𝑖𝑡 stands for the number of rooms at a certain property on time 𝑡. T. 𝑡𝑖 stands for the type of property and which has four categories. For this variable, 0.0895 is the coefficient for terraced properties, 0.148 for semi-detached, and 0.356 for detached. In order to adjust for perfect multicollinearity, the coefficient for an apartment is not included. 𝑔𝑖𝑡 indicates whether or not the property features a garage. 𝑐𝑖𝑡 provides the coefficient for central heating. 𝑙𝑖𝑡 is the dummy
variable for the status of the building, is it listed (1) or not (0). 𝑏𝑖 is the construction year dummy. All the properties that were built before 1945 do not have a dummy to adjust for perfect multicollinearity. Table 2 provides the coefficients for all the construction year dummies.
Construction year
Coefficient
Construction year 1945-1959
-.019
Construction year 1960-1970
-.022
Construction year 1971-1980
.030
Construction year 1981-1990
.114
Construction year 1991-2000
.206
Construction year ≥ 2001
.269
20 However, in their model they incorporate the yearly fixed effects. For the year 2014, this coefficient is unknown. In order to optimize the model, the yearly fixed effects of 2014 have to be incorporated. Due to the fact that Koster & Van Ommeren (2015) do not incorporate the year 2014 in their research this has to be estimated. The OLS equation that is used in their research has as requirement that the squared error term is minimized. In order to estimate the yearly fixed effects, linear programming is used. The equation is optimized such that the squared error term is minimized. In this equation the yearly fixed effect is the unknown variable. Based on this calculation the yearly fixed effect for 2014 is estimated, which is a year fixed effect of 0.689 for all the properties sold in 2014.
21 Property data
The dataset for the property value is provided by the NVM and covers 70% of all the property transactions made in the Province of Groningen in the year 2014. In total the database contained 4851 properties that were sold in the province of Groningen in year. For every transaction the realtors record a total of 69 characteristics. For this analysis the same cut-off values will be used as in the research of Koster and van Ommeren (2015). This means that the minimum price per m2 should be between 500 and 5000 euros. The size of the property is also subjected to a limitation. The squared meters per property should be between the 30 and 250. This is done to exclude extreme outliers that heavily affect the estimation. Furthermore for a few properties the postcode was missing; since the identification of earthquake exposure is based on postcodes, these were also removed from the dataset. Properties that are characterized as building plots or garage are removed. After this selection a total 4301 transactions were used for this analysis. Table 3 shows the characteristics of the dataset. Compared with the research of Koster and van Ommeren the property price per square meter is on average higher (1365), this can be explained because they did not already correct for yearly effects in their descriptive statistics (i.e. inflation and increase in property value). It is remarkable that in 2014 the share of detached properties has grown compared with the period from 1996-2013. The transition of this share was mainly provided by the semi-detached properties, which declined from 26% to 16.7%.
mean sd min. max.
Property price per M2
1513.4
461.8
500
3541
Number of earthquakes (PGV>0.5cm/s)
0.380
1.343
0
15
Size of property
107.81
35.69
30
250
Number of rooms
4.369
1.304
1
13
Type of property - Apartment
0.305
-
0
1
Type of property - Terraced
0.221
-
0
1
Type of property - Semi detached
0.167
-
0
1
Type of property - Detached
0.307
-
0
1
Garage
0.341
-
0
1
Central heating
0.967
-
0
1
Listed building
0.008
-
0
1
Construction year ≤ 1944
0.306
-
0
1
Construction year 1945-1959
0.069
-
0
1
Construction year 1960-1970
0.171
-
0
1
Construction year 1971-1980
0.161
-
0
1
Construction year 1981-1990
0.089
-
0
1
Construction year 1991-2000
0.113
-
0
1
Construction year ≥ 2001
0.091
-
0
1
22 Geological data
For this research it is essential to compare properties that were exposed to earthquakes with a PGV higher than 0.5cm/s and properties that lie outside those areas. The KNMI collects all earthquake data in the Netherlands with a sophisticated network of seismographs. Using the KNMI database on seismological activity it is possible to subtract all the earthquakes since 1991 (KNMI, 2015). In order to calculate which earthquakes where strong enough to cause damage to properties the equations 2.1 and 2.2 are used. The formula to calculate the peak ground velocity is used to calculate the PGV of every earthquake at a certain postcode.
log10𝑣𝑖𝑡 = −1.53 + 0.74𝑚𝐿𝑗𝑡− 1.33 log10𝑟𝑖𝑗𝑡− 0.00139𝑟𝑖𝑗𝑡 (2.1)
𝑟𝑖𝑗𝑡= √𝑑𝑖𝑗𝑡2 + 𝑠
𝑖𝑗𝑡2 (2.2)
vit represents the peak ground velocity in cm/s at a certain location. mLjt is the magnitude of a certain
earthquake in an certain year. The rijt is calculated with equation2.2. In this formula 𝑑𝑖𝑗𝑡2 is the distance
in kilometres between location i and the epicentre, location j. 𝑠𝑖𝑗𝑡2 is the source depth of the earthquake, this point is called the hypocentre. Based on the research of Koster and van Ommeren (2015), an average depth of 2 km is used for locating the hypocentre. The epicentre is the point where the earthquake is at its strongest at the surface of the earth. For every earthquake stronger than 2.2 of magnitude, the PGV at all postcode locations are calculated. For every earthquake the postcode areas that are in the zone of a PGV higher than the minimum level of 0.5 cm/s are counted. The 0.5cm/s is set by the research of Wu et al. (2004). For every earthquake in the Province of Groningen since 1991 the PGV is calculated. For a PGV > 0.5cm/s at the epicentre the magnitude has to be at least 2.2.
23 Table 4: The number of earthquakes with a PGV>0.5cm/s and their location, magnitude and radius in 2014.
Based on this information the earthquake areas were located. Figure 2 shows the locations on a map and their radius. The areas within the circles are identified as earthquake areas. There are two areas where there is a clear overlap. This means that the properties in this overlap were exposed twice to an earthquake with a PGV higher can 0.5cm/s in the year 2014.
number location magnitude PGV> 0.5cm/s radius in KM
Earthquake 1 Woudbloem 2.8 3.76 Earthquake 2 Zandeweerd 2.9 4.40 Earthquake 3 Garmerwolde 2.8 3.76 Earthquake 4 Froombosch 2.6 2.63 Earthquake 5 Schildwolde 2.3 1.04 Earthquake 6 Leermens 3 5.11
24 Test Results
After the analyses of the property data and geological data the prediction formula (1.1) is used to forecast the transaction price of every property sold in the Province of Groningen in 2014. This chapter will first analyse the outcomes on a price per m2 level and after this the price per property will be analysed. The reason for this is that in the research of Koster and van Ommeren (2015), they use the m2 price. However, for a monitoring system and usable output the model has to work on a property level. Using the property data and geological data, the model is employed to forecast the expected natural log price per m2 for the properties sold in 2014. The forecast is done in Excel and uses the NVM characteristics of all the properties sold in 2014. For the number of earthquakes the sum of earthquakes per postcode is counted from 1991. As can be seen in the geological data, there were six earthquakes with a PGV higher than 0.5 cm/s in 2014. Graph 1 shows the subdivision of all the properties sold in 2014 that were exposed to at least one earthquake since 1991. There is even a property that was exposed to 15 earthquakes with a PGV higher than 0.5cm/s.
Graph 1: The subdivision of properties that were exposed to at least one earthquake between 1991-2014 in the Province of Groningen and were sold in 2014.
228 94 119 33 17 10 22 11 12 6 11 2 0 1 1 0 50 100 150 200 250 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Num ber o f pro pert ies so ld in 2 0 1 4
number of earthquakes since 1991 (PGV>0.5cm/s)
Number of properties sold
25 However, some researchers have pointed out that the PGV is not the right way to measure the impact of an earthquake. It only collects the number of earthquakes that have a PGV> 0.5cm/s. The problem with the method used to calculate the influence of the earthquake is that there is only a selection on the basis of the extent in which the inhabitants can feel the earthquake. An earthquake with a magnitude 3.6 can cause more severe damage than a magnitude 2.3 earthquake.
Based on the measurements of Koster and van Ommeren (2015) the model forecasts the transaction value per m2 for every property sold in 2014. In total 4301 transactions are used to see to which extent the model was able to capture the property value. All this was done to validate the proposed appraisal method. In the end, this results in the validation of the expressed model that works towards an individual valuation. To validate the model the squared should be as high as possible. The R-squared measures the fit of the model. In this case it measures to which extend the model is able to forecast the historical transaction value. In the research of Koster and van Ommeren (2015) they use the OLS regression, which fits a line that minimizes the squared errors. In the model for the appraisal method this error is the difference between the real transaction price in 2014 and the forecasted transaction price following the coefficients of Koster and van Ommeren (2015). The smaller the difference between the forecasted values and the historical transaction value is, the higher the R-squared will be. Furthermore, the mean of the error terms has to be equal to zero; this is an obligatory assumption of the OLS (Gauss-Markov assumption). The standard error of the forecasted natural log price per M2 is 0.165. The standard error of the historical natural log price per M2 is 0.317. In order to calculate to which extend the model is able to forecast the property value the R-squared is used. Based on the difference between the forecasted price per m2 and the historical price per m2 the R-squared is calculated. For this the following formula is used:
𝑅2= 1 −∑(𝑦𝑖−𝑓𝑖)2
∑(𝑦𝑖−ӯ)2 (3.1)
In formula 3.1 the R2 is the fit of the model, 𝑦𝑖 the historical transaction value per m2 of property i, 𝑓𝑖
26 effects could improve the estimation even more. In the research of Koster and van Ommeren (2015) a total of 3,733 postcode fixed effects were used. Those missing postcode fixed effects cancel each other out, so it is still valid to use the model. Graph 2 shows the subdivision of the prediction error terms.
Graph 2: The prediction error of model per m2 for all the properties sold in 2014.
As can be seen in the graph, the model estimates the transaction price most of the time close or really close to the real transaction price. However, the outliers should be minimalized in an improved model. It could be possible that the postcode dummies are able to capture a large part of those outliers. However, the difference between the R-squared of Koster and van Ommeren (2015) and the R-squared of the proposed model is 7 percentage points, which means that including postcode dummies would improve the model but this is not the sole answer to reduce the outliers. To reduce the remaining prediction error, future research should include more property characteristics.
The NVM database has a rich and well-completed range of property characteristics. Based on the research that is performed it would be recommended for further research to see to which extend the following characteristics are able to improve the model. First the square meters of land, in the current model only the property size is used in the model. However, in parts of the Province of Groningen, most of the properties have large plot sizes. This increases the value and if this effect is included in the model some outliers could be removed. An analysis on the model shows that on average a property has 451 m2 of land but with a standard deviation of 1375 m2. Furthermore, the properties that were severe undervalued (<-0.75 ln of) had an average land size of 952 m2. This indicates that the model could be improved by including the square meters of land. However, this has to be tested with a hedonic pricing model to determine the effect and its signification. Another characteristic that could improve the model is the depth of the earthquake. In the research of Koster and van Ommeren (2015) they use an average depth of 2 km. However, this depth is not always the same. The KNMI also provides the depth of the earthquake i.e. the hypocentre. For a better method this should be taken into account before the
0 50 100 150 200 -2 3 0 0 -2 1 0 0 -1 9 0 0 -1 7 0 0 -1 5 0 0 -1 3 0 0 -1 1 0 0 -9 0 0 -7 0 0 -5 0 0 -3 0 0 -1 0 0 1 0 0 3 0 0 5 0 0 7 0 0 9 0 0 1 1 0 0 1 3 0 0 1 5 0 0 1 7 0 0 F re que nt io n
Difference between forecasted and historical value
Prediction error of the model
27 hedonic pricing model is estimated. Besides of this the model does not adapt to the soil type. Research has shown that the soil has a great influence on the damage on a property (Wassing & Dost, 2012). The soil in the Province of Groningen differs from clay to peat land and sand. All those ground types act different and cause different influences on the properties above it. This does not take away the influence of the soil type. Further research should focus to which extend the soil type influences the depreciation of property value when gas extraction induced earthquakes occur.
To come back to the core of this research the model should capture the influence of the earthquakes on the property value. In order to test this, the model was run with and without earthquake effects for all properties that were exposed to at least one earthquake since 1991. Table 5 shows the mean error term, variance and total number of observation from this test.
Table 5: Influence of earthquake effect
The values are made absolute to remove the smoothening effect of negative and positive values. For testing the outcomes the two-sample t-test assuming unequal variances is used to test whether the following hypotheses hold, in which the earthquake effect is the decrease in property value due to earthquakes.
H0: The Earthquake effect does not explain part of the property value H1: The Earthquake effect does explain part of the property value
The t-test showed that the earthquake effect significantly improves the model (P<0.001) with an alpha of 0.05. This implies that there is a significant effect of earthquakes on the property value in 2014. This is in line with the findings of Koster and van Ommeren (2015). The next step for validation the model is to make the switch from the value per m2 to the total property value. The model that is used in this research uses the price per m2 as basis. However, for property owners it would be more helpful to the extent of the loss in property value. In the next part this step will be made. For a well-established monitor system, which is eventually the goal, it is necessary that the model also works on a property level.
With earthquake effects Without earthquake effects
Mean error 0.115 0.175
Variance 0.068 0.073
28 From m2 to total property value
The model used in this research forecast the value per m2. However, to generalize the results towards a property level model the transaction price per property is calculated. This is done to provide a better overview on the difference between the forecasted price and the real transaction price. The tests before showed that the model had a R-squared of 74.2%. Furthermore, it showed that there was a significant negative effect of the earthquakes on the property value. The next step of this research is to generate the total damage to property value and to which extend the model is able to forecast this. When the model is used to calculate the total property value (price per m2 * property size) the R-squared is 46%. The reason why this is lower is due to the fact that the multiple property size differs the weight of the prediction error. For example, an m2 price error of 200 euro has a different influence for a total value of a property with a property size of 100m2 compared with a property of 200m2. The lower R-squared provides reasons for further research to use the price of the property instead of the price per m2 as dependent variable. However, this is not part of the research.
For the compensation programme the most important part is the influence of the earthquakes on the property price. In order to address this, all the properties that were sold in 2014 and which suffered from at least one earthquake whit a PGV > 0.5cm/s are selected. This selection is made because the focus is now on the total damage due to the earthquakes. For this selection, the average damage to the property value is 3787.08 euro per earthquake with a PGV>0.5cm/s. Graph 3 shows the average damage per earthquake for each subdivision on the primary as and the total average damage per subdivision on the secondary as. This subdivision is the number of earthquakes a property is exposed to.
Graph 3: The average damage per earthquake and the total damage to the property value. 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 1 2 3 4 5 6 7 8 9 10 11 12 14 15 0 10000 20000 30000 40000 50000 60000 a v er a g e da m a g e per ea rt hq ua k e
Number of earthquakes since 1991
To ta l da m a g e to pro per ty v a lue
Influence of the Earthquakes on property value
Average damage per earthquake
29 Based on the estimation error all the properties that were exposed to at least one earthquake since 1991 are selected. To test if there is still an earthquake effect the following hypothesis is tested with a t-test with unequal variances, table 6 shows the characteristics of the two samples.
The mean prediction error of the forecast with earthquake effects is remarkable smaller. The results of the t-test show that adding the earthquake effects has a significant positive influence on the error term (P<0.0001), which can be interpreted that there is a significant influence of earthquakes on the property value. This is in line with the results from the model, which forecasts the price per m2. So, to conclude this result section: on average the model is able to forecast value, on a per square meter level and on the whole property level. This implies that the proposed appraisal method in this research is usable for the valuation of properties in the Province of Groningen. With the model, it is now possible to construct a monitoring system that can forecast the property value, and above all, the damage due to earthquakes on the property value. This will provide the property owners with more information about their situation. Secondly, the municipalities are now able to calculate the decrease in OZB value due to earthquakes. The proposed appraisal method is validated. However, as mentioned before there is still room for improvements.
With earthquake effects Without earthquake effects
Mean prediction error 12,300.19 23,583.94
Observations 567 567
30
Conclusion
31
References
Bakker, M., 2015. The winning of Natural Gas and Its Effect on Real Estate Value: Exploring the Design of a new monitoring system. Master thesis Technology and Operations Management, University of Groningen
Balsters, H., 2013. Data & Process Integration: Deriving Business Data from Business Processes,
with an application to Data Management in Energy, Groningen, The Netherlands: University of
Groningen.
Balsters, H., 2015. Design Methods: Building utility-driven Artifacts. Groningen, The Netherlands.
Dagblad van het Noorden, 2015. NAM: nu compenseren is volstrekte willekeur. Available at:
http://www.dvhn.nl/nieuws/nederland/nam-nu-compenseren-is-volstrekte-willekeur
De Kam, G.R.W. & Elhorst, J.P., 2014. Inzet RUG voor onderzoek naar impact aardbevingen op
wonen en welbevinden, Groningen.
Droës M.I. & Koster R.A.H., 2014. Renewable Energy and Negative Externalities: The effect of Wind turbines on house prices, Tinbergen Institute discussion paper, 124(8).
Francke, M.K. & Lee, K.M., 2013. De waardeontwikkeling op de woningmarkt in aardbevingsgevoelige gebieden rond het Groningenveld, Rotterdam, The Netherlands.
Hagoort, J., 2015. Aardbevingen in Groningen, Verleden en Toekomst. Available at:
www.deingenieur.nl/aardbevingen
Knight, J.R., 2002. Listing price, time on the market, and ultimate selling price: Causes and effects of listing price changes. Real estate economics, 30(2), 213-237.
KNMI, 2015. Geïnduceerde aardbevingen in Nederland. knmi.nl, p.20-21. Available at:
32 Koster, H.R.A. & Van Ommeren, J., 2015. Natural Gas Extraction, Earthquakes and House Prices. Tinbergen Institue Discussion Paper, 38(8).
NAM, 2015. Feiten en cijfers. Available at: http://feitenencijfers.namplatform.nl/waarderegeling/
NRC, 2015. Raad van State zet voorlopig streep door gaswinning Loppersum. Available at:
http://www.nrc.nl/nieuws/2015/04/14/raad-van-state-zet-voorlopig-streep-door-gaswinning-loppersum/
Malpezzi, S., Ozanne L. & Thiodeau, T., 1980. Characteristics Prices of housing in 59 SMSAs. The Urban Institute. Washington, DC.
Monson, M., 2009. Valuation using hedonic pricing models. Cornell Real Estate Review, 7, 62-73.
Porter, K.A., Beck, J.L., Shaikhutdinov, R.V., Au, S.K., Mizukoshi, K., Miyamura, M., Ishida, H., Moroi, T., Tsukada, Y. & Masuda, M., 2004. “Effect of seismic risk on lifetime property value.” Earthquake Spectra, 20(4), 1211-1237.
Sirmans, G.S., MacDonald L. & Macpherson, D.A., 2010. A Meta-Analysis of selling price and time-on-the-market. Journal of housing research. 19(2): 139-152.
Sirmans, G. S., Macpherson, D. A. & Zietz, E. N.b 2005. The Composition of Hedonic Pricing Models. Journal Of Real Estate Literature, 13(1): 3-43.
Taylor, C., 1999. Time-on-the-market as a sign of quality. Review of economic studies. 66: 555-578.
Van Strien, P.J., 1997. Towards a Methodology of Psychological Practice: The Regulative Cycle. Theory of Psychology, 7(5): 683-700.
Volkskrant, 2015. Bewijslast bevingsschade wordt omgedraaid. Available at:
http://www.volkskrant.nl/politiek/bewijslast-bevingsschade-wordt-omgedraaid~a3986959/
33 Wieringa, R., 2007. Writing a Report About Design Research, Enschede, The Netherlands.
34
Appendix 1 Stakeholders, Goals and CSF’s
Stakeholder Goals CSF’s functional CSF’s non-functional
Ministry of economic affairs
Use a calculation that uses multiple calculation methods (second opinion).
Increase knowledge of effects of winning of gas on real estate value.
The system shall use multiple calculation methods and shall deliver knowledge increasing information.
The system shall be flexible enough to incorporate multiple calculation methods.
NAM Use a calculation that uses multiple calculation methods (second opinion).
The system shall use multiple calculation methods.
The system shall be flexible enough to incorporate multiple calculation methods.
Province of Groningen
Use a calculation that uses multiple calculation methods (second opinion).
The system shall use multiple calculation methods and shall output a grounded, validated value for depreciation.
The system shall be flexible enough to incorporate multiple calculation methods.
Municipality Use a calculation that uses multiple calculation methods (second opinion) and maximises depreciation.
The system shall use multiple calculation methods and shall maximise the calculated depreciation.
The system shall be flexible enough to incorporate multiple calculation methods.
Residents – VEH
Use a calculation that uses multiple calculation methods (second opinion) and maximises depreciation.
Should bring security to housing owners in the risk area.
The system shall use multiple calculation methods and shall maximise the calculated depreciation. It shall output a grounded, validated value for depreciation.
The system shall be flexible enough to incorporate multiple calculation methods. It shall be accepted by all stakeholders thus brining security to housing owners.
Residents – GBB
Use a calculation that uses multiple calculation methods (second opinion) and maximises depreciation.
Should bring security to housing owners in the risk area.
The system shall use multiple calculation methods and maximise the calculated
depreciation. It shall output a grounded, validated value for depreciation.
The system shall be flexible enough to incorporate multiple calculation methods. It shall be accepted by all stakeholders thus brining security to housing owners.
All Have a system that: Outputs usable data Outputs complete data Is reliable
Needs low maintenance Is flexible to changing context
The system shall bundle data from multiple sources to output usable, complete data for its users.
The system shall be reliable in terms of uptime, shall need low maintenance in terms of reducing the number of times data has to be entered
manually and the system shall be flexible enough to adapt to a changing context.
36
Appendix 3: Proposed appraisal method by Bakker (2015)
37
