Name: Thomas Wissink
Student number: 5979447
Telephone: +316 50513835
Email: t.p.wissink@student.tue.nl
Discipline: Master Business Economics Graduation committee: Dr. M. Theebe
Prof. dr. M. Francke (2nd corrector)
Photovoltaic system deployment by
dwelling-owners as a function of alternative
investments performance –
A discrete choice analysis.
3 Contents
1. Introduction of Photovoltaic system investing ... 7
1.1 Introduction in PV systems ... 7
1.2 Problem... 8
1.3 Problem owners ... 8
1.4 Goal ... 9
1.5 Main research question ... 9
1.6 Hypotheses ... 9
1.7 Sub questions... 9
1.8 Conceptual model ... 9
1.9 Expected result ... 9
1.10 Further composition of the report ... 9
2. Selection of the methodology ... 11
2.1 Existing Literature ... 11
2.2 Modeling PV investing ... 12
2.3 Introducing the discrete choice experiment ... 14
3. The design of the discrete choice experiment ... 17
3.1 Theory underlying discrete choice ... 17
3.2 Attributes ... 18
3.2.1 Stock performance ... 20
3.2.2 Deposit returns ... 21
3.2.3 Mortgage pay-off ... 21
3.2.4 Generic attributes ... 21
3.3 Design of the experiment ... 21
4. Conduction of the experiment ... 27
4.1 Data collection ... 27
4.2 Model fit ... 28
4.3 Binomial logit model estimation ... 28
4.4 Descriptive statistics ... 31
4.5 Cross-elasticities and graphic display of results. ... 35
5. Conclusions ... 39
5.1 Introduction ... 39
4
5.3 Further findings ... 40
5.4 To conclude ... 41
6. Discussion & recommendations ... 43
Literature ... 44 Appendices ... 47 Appendix A ... 47 Appendix B ... 48 Appendix C ... 49 Appendix D... 50 Appendix E ... 51 Appendix F ... 52 Appendix G ... 53 Appendix H... 54 Appendix I ... 55 Appendix J ... 56 Appendix K ... 57 Appendix L ... 58 Appendix M ... 59 Appendix N ... 60 Appendix O ... 61 Appendix P ... 62 Appendix Q ... 63 Appendix R ... 65
7
1. Introduction of Photovoltaic system investing
In this chapter the subject is introduced. Secondly the practical relevance of the research is explained. Lastly the first outline is made of the research design.
1.1 Introduction in PV systems
There is a transfer going on in the western world to a more sustainable footprint of human activities. Or at least a lot of people and companies are engaged in the transfer to a more sustainable footprint. The Netherlands are no exception. Since the built environment has a large share in energy use, all sorts of measures that increase energy efficiency have been invented for buildings. Energy efficiency has become an important dwelling attribute next to floor space, location and deterioration. Most new buildings are very energy efficient. A problem is however that the building stock is not being renewed fast. Energy efficiency in the built environment is therefore focusing on both the design of new energy efficient buildings and on the improvement of energy efficiency in the existing stock. (AgentschapNL, 2012)
Photovoltaic systems, or PV systems, are an energy saving solution that is easy to integrate with this existing building stock. PV systems generate electric current from energy of the sun in a way it can be used in the socket (EPIA, 2010). Other add-on measurements such as isolation or condensing boilers only limit energy use. Photovoltaic cells can change buildings from energy users into energy producers. Therefore it is not surprising that Photovoltaic cells are broadly deployed in the built environment in the Netherlands. That is, the growth of deployment is large. The energy produced by PV cells, as share of the total Dutch energy use, including transport, industry and households, is only 0,038% (CBS, 2012). But it has to be said that the Netherlands are far behind other countries in Europe: when ranking European PV generation as a part of the total use they take a seventeenth place. Germany and Spain, for example, generate 0,75% of total use. However, the Dutch growth in PV power generated is not low: 64% growth in from 2010 to 2011, (CBS, 2012) and 138% from 2011 to 2012 (Cobouw, 2013). The European Renewable Energy Council expects PV generation percentages for Europe to be 1,3% in 2020 and 11% in 2050 (EREC, 2010).
In 2012 Dutch dwellings owners invested over €200.000.000 in PV technology (Cobouw, 2013). PV technology is a decent investment alternative for households in possession of a dwelling. Naturally a PV system is not completely comparable with other retail investment possibilities. Unlike these a PV system can only be bought once. Also unlike other investments selling it again is difficult. Lastly the PV technology and PV legislation when investing today may have developed a lot tomorrow, as PV is still in great motion.
But this does not mean that at present time the conditions for investing in PV technology are not good; sustainability is a popular concept and traditional investment alternatives such as deposits and stock are in heavy weather. Also amortizing mortgage is not always an optimal choice from a financial perspective due to Dutch tax legislation and low mortgage interest rates. The returns on deposits, stock and bonds are historically low.
Will the amount of investment in PV be as large as it is now when this situation changes and the interest rates and stock prices are increasing? Do retail investors see PV systems as just another investment alternative with merely its financial performance? Or are there other
8
factors that play a role such as ideal motives and/or immaterial revenue, and is it a class on its own?
Existing research has proven that the most important reasons for investing in PV technology are not only of financial nature. Research of Sormani finds that 37% of investors have idealistic motives (Sormani, 2011). Wissink found that 17% have primarily idealistic motives and 27% have less dependency of energy prices as primary motive (Wissink, 2013). Also the “fun of the pioneering” is often heard (Profinergy, 2013). If for certain people next to the financial performance also other non-monetary returns exist for PV investing, this could mean that these people are less sensitive to changes in the financial performance of PV systems relative to other investment alternatives. This might not apply to investors that see PV systems merely as a financial vehicle. This means there could be different groups with different behavior regarding PV investing. The question is if this difference does indeed exist and how large it is.
Another interesting question is which behavior should be measured: the decision behavior of people that already have a PV system (would they buy it again?) or the behavior of those that might invest in a PV system in the future (would they buy it?). Not only do both alternatives have their own advantages, but the difference between both groups may also tell us a lot about the changing opinion about new products when one becomes more acquainted with it relative to having no experience with the product.
1.2 Problem
Photovoltaic panel deployment in owner occupied housing is growing steadily against the background of an economic crisis. The low interest rates and unpredictable stock performance in the Netherlands as a result of this crisis are expected to make investing in PV technology on owner occupied dwellings relatively more attractive. This may have a huge effect on the PV business if the economic landscape changes. It is however not known to what extent the choice to invest in PV systems is dependent on the performance other investment alternatives.
1.3 Problem owners
There are four problem owners:
Companies that are involved in installing or selling PV panels.
The change in demand for their products when other investment alternatives become more attractive is essential to their revenue. To what extent they compete with other investment alternatives is valuable information.
Energy companies.
These companies are sharply following the developments in the dynamic energy market. They are struggling with their changing role. (Bearingpoint, 2012).
Policy makers.
The energy markets in general and the market for PV systems in specific are very dependent on policy, subsidies and taxes. Policy has a large influence on deployment, but deployment of PV systems also influences policy.
9
How investing in PV systems by retail investors is relating to the performance of other investment alternatives is interesting for this group since they are dependent of retail investment.
1.4 Goal
To gain insight in the behavior of investors regarding PV systems as a function of the hypothetical future performance of other investment alternatives through a survey that should result in a choice prediction model.
1.5 Main research question
What is the relation between traditional retail investment performance and PV deployment? 1.6 Hypotheses
•The choice whether to invest in a PV system or not is influenced by the performance of other investment opportunities available to households.
•There is a difference between the choice behavior of people that already have a PV system and the ones that do not.
•People that perceive immaterial returns of a PV system are less sensitive to increases in financial returns of alternative investment classes.
1.7 Sub questions
•What are the most important alternatives to invest in for retail investors that generate varying return?
•What are the most important factors that influence the choice of these alternatives? And how can those factors be controlled for?
•What are the effects of changes in expected returns of alternatives on PV investment? 1.8 Conceptual model
vgffvv
Figure 1.1. Unit of research: ground based dwelling owners with- and without a PV system.
1.9 Expected result
The result will be an elasticity between a change in return of a certain investment type and a change in the percentage of dwelling-owners willing to invest in PV systems. This elasticity may be non-linear. The elasticity will be calculated for different relevant groups.
1.10 Further composition of the report
In chapter 2 the literature is elaborated on and consequently the method for the research to reach the goal is chosen. In chapter 3 this method and its theory is explained, applied on the case and an experiment is designed. In chapter 4 the experiment in the form of a survey is conducted and the results are reported. In chapter 5 the conclusions are drawn. Chapter 6 reflects on the results and puts them in a societal and practical context; also recommendations for further research are made.
Performance of traditional investment options accessible by dwelling owners. PV panel deployment by dwelling owners.
11
2. Selection of the methodology
In this chapter it is explained which methods can be helpful to answer the research questions stated in chapter 1. Firstly existing research that can be useful is discussed. Secondly the specific case of this research is introduced. Lastly a method is chosen and explained.
2.1 Existing Literature
A lot of research has been done into the behavior of retail investors all over the world towards different investment alternatives (Mankiw, 1991; Hochguertel et al., 1997; Bertaut and Haliassos, 1997; Flavas, 1998; Alessie, 1999; Odeon, 1999; Barber and Odeon, 2000; Letendre, 2001; Chapman, 2005; Cocco, 2005; Dimmock, 2010; Bateman, 2010, Jin, 2011). This research shows that household investment behavior is not always optimal from a rational portfolio perspective. This makes it harder to make predictions about PV investment behavior of households. The existing research also uses all kinds of household investment class divisions. These classes are displayed in table 3.1 in the next chapter.
All of the mentioned research uses revealed preference data. That means that no real behavior is observed, but people is asked how they behave, or how they would behave in a hypothetical situation. The reason for this is that true portfolios of households are confidential and not easy to map. Some studies with cooperation of brokerage houses do however exist (Investment Company institute, 2000; 2002). A yearly conducted survey by the Dutch central bank, “De Nederlandsche bank” is the leading survey in monitoring Dutch retail investment behavior, more on this will be explained in chapter 3. Surveying is also required in this type of research, because the research intends to explore behavior in a hypothetical future situation.
In the literature about Households investments behavior Mankiw introduces the equity premium puzzle. This is a concept introduced in 1985 by Rajnish Mehra and Edward C. Prescott. The equity premium puzzle states that it is inexplicable that households hold few stock, taking into account the higher expected returns. The equity premium puzzle states that the higher risk for stock is an insufficient explanation for this. Mankiw tries to solve the puzzle by stating that households could be divided into stockholders and nonstockholders. If this is done the amount of stock held by the stockholders is a bit more as to be expected from a rational perspective, but in his American dataset it remains unexplained why quite some very wealthy people do not hold stock.
Mankiw proves that the choices of households regarding finance are not always optimal due to extreme risk aversion of some households. Dimmock concludes the same from a Dutch data set. He calls this extreme risk aversion “loss aversion”. Jin includes the private business and leverage in the investment portfolio of his US household data and states that the risk coming from this helps to explain the sub optimal stock holdings especially for young households. Alessie finds that in that more and more Dutch households may have started investing in stock, funds, innovative mortgages and so on in the nineties, but that a large share of households is still very traditional and puts all savings in to deposits accounts and mortgage amortization. Alessie also claims that retail investors in the Netherlands are
12
sensitive for returns. If the returns of certain investment classes go up, then, ceteris paribus, people tend to allocate money to that class.
It was proven by Chapman, Barber and Odeon and Odeon that households in the United States have the same tendency of shifting investments to stocks when the returns of the S&P 500 increase and vice versa.
Hochguertel et al. shows that in the Netherlands risk averse households do not diversify their investments even though this decreases risk.
Bertaut and Haliassos also state that stockholding is lower than expected and they find that higher educated households do on average have a longer investment horizon.
Flavas has done research in the US and argues that due to owner occupation of dwellings households invest more in real estate than would be rational from a portfolio diversification perspective. He also claims that young households own relatively few equity because they often have high leverage and their prime focus is on mortgage amortization. Cocco on the other hand found a correlation in US data between mortgage debt and stockholding and concludes that it is likely that households finance stock with mortgage debt.
Letendre shows from US data that there is a relation between income risk and the amount of risk a household is willing to take with its investment portfolio. On average he finds; the more income risk the less investment risk households are willing to take, leading to sub optimal portfolios.
This research mostly shows us that the investments behavior of households is often sub-optimal and hard to predict because a lot of personal factors play a role. This means that it will be important to carefully gather demographic information about respondents.
The existing research above focuses only on traditional investments. PV panels have not been researched as an investment class so far. Sormani (2011) researched the preferences of retail investors regarding certain attributes of PV systems such as payback period and quality. But the influence of other investment alternatives was not controlled for in his research.
It is interesting to look into the used methods of the existing research into the behavior of retail investors. The method of this research could be used by adding a PV system as an alternative. Methods often used are: the discrete choice method (Alessie, 1999, Dimmock, 2010; Bateman, 2010; Kitamura, 2012) and direct surveying (Mankiw, 1999; Haliassos, 1997; Flavin, 1998; Chapman, 2005; Cocco)
2.2 Modeling PV investing
Direct surveying means asking respondents directly what their behavior would be. It has some drawbacks (Breidert, 2006):
Directly asking opinions about unfamiliar situations is cognitively challenging for respondents and may lead to bias.
13
Research has showed that directly asking leads to unstable answers that can change abruptly without any particular reason.
Direct surveying is limited in the measurement of trade-off effects. Trade-off effects are the relative importance respondents attribute to things that have effect on their behavior.
With indirect surveying questions are asked and the factors of interest are derived from the answers. It is very suitable for calculating cross-elasticities in an efficient way without too much challenging of the respondents. Because of these reasons an indirect survey is chosen to estimate behavior regarding PV investing. However also direct surveying is done to gather demographic information about the respondents. This personal information may have a relation with choices made in the indirect survey.
Indirect data gathering has two suitable sub-methods; discrete choice measurement and conjoint measurement. Both methods construct hypothetical options with varying attributes and present these options to respondents in order to gather information about preferences. In a discrete choice experiment (DCE) respondents choose one option out of a selection of options. In conjoint measurement (CM) instead of choosing options, options are ranked or rated. And where conjoint measurement has its roots in marketing, analysis is purely mathematical and is called conjoint analysis (CA). The analysis of discrete choice measurement is called discrete choice analysis (DCA), the analysis relies more on micro-economic theory (Hensher et al., 2005). Results have not proved to differ in accuracy between both alternatives but DCA has more possibilities (Breidert, 2006; Louviere, 1994). Because of this in combination with the successful discrete choice experiments in the literature discrete choice has been chosen as the method for data gathering and data analysis. Figure 2.1 shows the process method selection resulting in direct surveying and discrete choice method.
Figure 2.1. The process of choosing the methodology of researching retail investors’ behavior regarding PV
investing as a function of other investment alternatives returns. Options that were not chosen are made transparent.
14 2.3 Introducing the discrete choice experiment
Through a discrete choice experiment with stated preference data, the effect on respondents’ decision to invest in PV panels is measured when the historic returns of other investment alternatives hypothetically changes. The respondent can choose to invest in a PV system or can choose not to invest in PV and thus to invest in another alternative automatically. Two types of respondents are surveyed. Respondents in possession of a PV panel should indicate whether they would also have invested in their PV system given the varying returns. Respondents in possession of a ground based dwelling but not of a PV system should imagine whether they would invest in a PV system if they had €7.000 saved after consumption. The reason that not only people that have already invested in PV are interviewed is that they might have embraced the PV system now that they have it: This could bias their choice. It could be that in reality these people would not invest in PV anymore given higher returns on other alternatives. The group without a PV system does not have this possible bias. But they have another possible bias: despite the fact that they indicated they would invest in a PV system if they had the time and money, it is not certain that they would in reality do so. In the next chapter it is elaborated how the experiment is designed exactly.
17
3. The design of the discrete choice experiment
In this chapter the experiment is designed through examining and combing what has been done in the literature and applying this on the PV case. The result will be all the preconditions to create an online survey. Firstly the theory behind the research is explained. Secondly the attributes and its levels are chosen and lastly a statistically efficient design of the experiment is made.
3.1 Theory underlying discrete choice
There are different discrete choice models. The dependent variable of this research is binary: a respondent chooses either to invest in a PV system or not. The response of a respondent can be either 1 or 0.
Means that respondent choose to invest in a PV system. Means that respondent choose not to invest in a PV system
Before proceeding it is necessary to introduce the principles of random utility theory (RUT). RUT assumes that all individuals when they are able to choose between alternatives, for example investing in PV or not, will choose the alternative with the highest utility. (Hensher et al., 2005). This is displayed in equation (1).
(1)
Where is the utility of the chosen alternative and is the other alternative in the
choice set that individual can choose.
RUT assumes that the utility of choosing to invest in a PV system exists of a systematic part that is explainable and a random part that is not explainable. (Hensher et al., 2005)
(2)
In equation (2) is the unobserved utility that an individual perceives from choosing to
invest in a PV system . is the systematic, explainable component and is the random
component. Because of the random component, the probability that an individual will choose to invest in a PV system given certain circumstances can be calculated, but the exact choice cannot be calculated.
Utility values of choice alternatives do not have any other meaning than relatively to the utilities of other choice alternatives, therefore the systematic utility of not choosing to invest in PV is set to zero:
(3)
The systematic component of choosing to invest in a PV system can be modeled as the sum of part- worth utilities that depend on the different situation attributes and their levels: in this case the returns of other investment possibilities. In other words the rates of the returns have a “part- effect” on the systematic utility .
18
Equation (4) states that the systematic utility of investing in PV exists of the sum of part-
worth utilities. is the value of attribute level of alternative that is in the choice set of
respondent . is a parameter that indicates the contribution of attribute on the utility of the alternative. Such an attribute could for example be the present interest rate on a savings account. (Hensher et al., 2005)
can be modeled to have a normal distribution or a logistic distribution. Both models do
not differ much but logistic choice modeling is more common unless there is strong evidence that the distribution of is normal. (Hensher et al., 2005)
With NLOGIT software it is possible to make estimations of . With these estimates the probability that a respondent will choose to invest in a PV system can be calculated. This probability is the e-power of the systematic component of investing in PV divided by the sum of the e- power of the systematic component of investing in PV and the e-power of the systematic utility of not investing in PV. The e-power of zero is one, see equation (5). (Hensher et al., 2005)
| (5)
With the changes in probability for different returns of other investment alternatives it is possible to calculate the cross-elasticities. This is done in chapter 4.5.
3.2 Attributes
The attributes of this discrete choice experiment are the rates of return of other investment alternatives. As explained in paragraph 2.1, a lot of research has been done into the behavior of retail investors all over the world towards different investment alternatives. Within this research investment alternatives are often divided into different classes. In table 3.1 these class-divisions are displayed for different articles from the literature.
DISTRIBUTION OF RETAIL INVESTMENT ALTERNATIVES Author and year No. Classes
Mankiw (1991) 2 Stock, Other liquid assets
Hochguertel (1997) 4 Risky assets (stock, bonds), Risk free assets (deposits ,saving accounts), Life insurences, Prime residence
Haliassos (1997) 2 risky assets (stock), Riskless assets (saving accounts, checking accounts, bonds, life insurances, CDs)
Flavin (1998) 5 Treasury bills (short term), Treasury bonds, Stock, House, Mortgage
Alessie (1999) 12 Bank accounts, Bonds, Stocks, Mutual funds, Defined contribution plans, Life insurance, Pension, House,-
Other real estate, Business equity, Durable goods, Other
Letendre (2001) 2 Risky assets, Riskless assets
Chapman (2005) 2 Stock, No-Stock
Cocco (2005) Stock, Cash, Bonds, Real estate, Vehicles, Real estate other than primary residence
Dimmock (2010) 2 Equity, No-Equity
Jin (2011) 4 Stock, Private Business, Real Estate, Riskless assets
Table 3.1. Distribution of retail investment alternatives in the literature.
The classes from table 3.1 show that there are a lot of ways to distribute investment classes. Also the perception of a risky asset seems to differ. Because this thesis tries to identify the
19
investment classes of which the returns have the highest expected influence on PV deployment it makes sense to look into the retail investment behavior of dwelling owners in the Netherlands. This behavior is displayed in table 3.2.
PERCENTAGE OF DUTCH DWELLINGOWNERS THAT POSSESSES A CERTAIN ASSET
Asset Liability
Checking account 97% Personal debt 2%
Corporate savings account 30% Revolving credit 10%
Savings account 90% Consumer credit 2%
Bankbook 2% Other debt 1%
Dutch "Koopsompolissen" 24% Debt with acquaintances 3%
Capital insurance 13% Study debt 3%
Mutual Funds 25% Credit card debt 3%
Bonds 5% Negative checking account 8%
Stock 14% Mortgage primary residence 65%
Options 1% Mortgage secondary residence 1%
Cars 83% Other mortgages 2%
Motorcycles 7% Boats 3% Caravans 14% Loans 10% Other assets 2% Substantial participations 1% Self-employed capital 5% Primary residence 100%
Life insurance for residence 17%
Second residence 4%
Other real estate 4%
Table 3.2. Percentage of Dutch dwellingowners that possesses a
certain asset (NIBUD, 2012)
The following things lead to the selection of attributes: Table 3.1.
Table 3.2.
Ease to invest in the alternative.
Limited correlation between the returns of alternatives.
The investment alternative should have a positive expected monetary return. This is for instance not the case with cars. They are expected to decrease in value (except for some old-timers). Their revenue is non-monetary: the use of the car.
These considerations led to the selection of three attributes. The possible assumed investment possibilities that may influence in PV are consequently:
Investing in stock. Returns are dividend plus stock price increases.
Putting money in a savings account covered by the “deposito garantie stelsel”. Paying off the mortgage. The return is the saved mortgage interest.
There are different ways to calculate returns. Returns can be corrected for inflation, taxation in box 3 and dividend tax. Since all investments are subject to inflation this is neglected.
20
Taxation in box 3 is also neglected because it is not equal to all households. Dividend tax is included since this is 15% for everyone. The exceptions on this, which applies to households with a substantial participation and some households abroad, are rare and thus neglected in this research. Lastly it is impossible to correct returns on mortgage payoff after tax since the tax deductibility differs among households. The return rates are set as follows:
Average yearly stock returns (price development + dividend) on the AEX index when invested five years ago, after dividend tax uncorrected for inflation and box 3 taxation.
The highest available interest on a savings account covered by the “deposito garantiestelsel” uncorrected for inflation and box 3 taxation.
Mortgage interest as charged by the bank.
These three alternatives are not completely comparable to investing in a PV system: A PV system can only be bought once, reselling it is difficult and it is a technology that is still developing fast just as the legislation about PV. Also the idealism that some investors put in PV is not seen as much with the other alternatives. These differences do not cause problems since it still remains interesting to study the effects of the performance of these alternatives on PV investing. These differences do however indeed play a role when trying to explain the effects.
3.2.1 Stock performance
Stock returns are hard to measure because it depends on the horizon. Because most retail investors tend to invest in national stock (Ahearne, 2004) it makes sense to use the AEX index for the perceived stock climate. Since the total return is measured after dividend tax, the net reinvestment index is used calculated by NYSE Euronext (ISIN code QS0011211156). The average yearly return over the last 5 years was 2,2% (√ ). And on top of that volatility was very high (figure 3.3).
Figure 3.3. The net AEX Total return index for the last 5 years (aex.nl, 2013).
The future attributes for what future PV investment behavior is measured is an average net AEX return of 2%, 6% and 10% for the last 5 years.
21 3.2.2 Deposit returns
Deposit returns are low when put in a historic perspective. The highest possible rate covered by the “deposito garantiestelsel” is as of 5-8-2013 1,95% (spaarrente.nl, 2013). The future attributes for what future PV investment behavior is measured is a variable interest of 2%, 4% and 6%.
3.2.3 Mortgage pay-off
Mortgage interest depends on the period over which the interest is fixed. Because mortgage interest is at a historic low the variable interest rate is the lowest. This variable rate will be use. Since the Rabobank is market leader and other big banks are not allowed to compete with lower prices since they received support by the Dutch government (NVM, 2013), the variable Mortgage interest at the Rabobank is 2,8% as of 5-8-2013 (Rabobank, 2013). The PV investment behavior is measured for a variable mortgage interest with the levels: 3%, 6% and 9%. Changes in the Dutch tax policy are not taken into account. Generic attributes state that tax deductibility remains. Certain respondents may be unable to pay off the mortgage because they have a certain type of mortgage or because they do not have one at all. This can be indicated in the Socio-demographic data gathering in the survey.
3.2.4 Generic attributes
Besides the attributes and attribute levels there is also need to mention circumstances that apply on all alternatives. These circumstances are important to mention because otherwise different respondents make different assumptions which will bias the perceived utility. These generic circumstances are:
The tax deductibility of mortgage interest remains as it is.
The hypothetical stock exchange behavior represents a past net return for the last five years.
It is assumed that no subsidies are present for PV systems.
The PV system presented to respondents is assumed to be a standardized system that costs €6000 and saves €600 per year for twenty years. Respondents are told that this is comparable to a 7,8% stock or deposit return. This is because it is assumed the panel has no value after 20 years, therefore depreciation decreases the return. These figures may lead to bias in favor of PV systems. After all, these values are expected values. Respondents may just accept the values as “true” because the survey-maker is the expert, there is however some risk. In reality things like energy price fluctuations and operating lives that deviate from 20 years may lead to other financial performance of the PV system.
3.3 Design of the experiment
Now that all attributes and attribute levels are defined the experiment can be designed. It should be noted that every additive piece of information or parameter that is measured will require an extra degree of freedom and thus an extra hypothetical scenario presented to the respondent (Hensher, 2005). It is important to limit this because otherwise respondents will lose their interest or will even quit. Not all possible scenarios (3 ) are presented but a statistically efficient design is made with the an Orthogonal array (Appendix A). An orthogonal array is a series of variations of attributes with zero correlation between these variations. In this way the effects on the dependent variable of the variations can be entirely attributed to one attribute. The attributes and their levels are displayed in table 3.4.
23
ATTRIBUTESXANDXATTRIBUTEXLEVELS
Attribute Attribute level Nr.bof levels Average net stock returns for the last five
years
2% 6% 10%
3
Max interest on savings account 2% 4% 6%
3
Variable mortgage interest 3% 6% 9%
3
Table 3.4. Attributes and attribute levels.
To prevent bias all levels should be represented to respondents an equal amount of times so the amount of scenarios generated should be a multiplication of three. The maximum acceptable times to represent a respondent with a hypothetical situation and let the respondent indicate whether he or she would invest in PV is set nine. The literature often mentions seven as maximum for choice sets for multinomial models (Brano, 2012). Multinomial models have more than two alternatives to choose from. This research uses a binomial model and thus nine seems acceptable.
The effects of the three investment alternative returns can be measured linear and non- linear. Because it is unknown what the relations between the attribute levels and choice are, it would be illegitimate to assume a linear relationship. Therefore non- linear effects are allowed in the model (Figure 3.5). This costs two degrees of freedom per attribute (one for each level minus one omitted level), which adds up to a total of six. Two degrees of freedom are needed for the intercept (β0) and the error term ( ). This results in the fact there is only
one degree of freedom left to model interaction effects for non-linear modeling. There are enough degrees of freedom left to model them all if the main effects are modeled linear. However, in Appendix A it shows that the correlation between all two- way interaction columns and design columns A, B and C are zero, but the interaction effects are 50% correlated between each other. Unfortunately the limited amount of scenarios for the sake of respondent dropout does lead to this correlation and does not allow to measure interaction effects in a proper way. With the data available it would be possible to model the interaction effects, but this model would suffer from severe multicollinearity. Nlogit ouput for this model is shown in Appendix B. This model has a bad model fit. Unfortunately interaction effects cannot be researched any further due to these practical problems.
24
The last step to finish the design is to assign the attributes and the attribute- levels to the design columns; this is done in table 3.6.
THE NINE SCENARIOS AS PRESENTED Treatment combination Stock return Deposit return Mortgage interest 1 2% 2% 3% 2 6% 4% 6% 3 10% 6% 9% 4 10% 2% 6% 5 2% 4% 9% 6 6% 6% 3% 7 6% 2% 9% 8 10% 4% 3% 9 2% 6% 6%
27
4. Conduction of the experiment
4.1 Data collection
The data is collected through an online questionnaire. Print screens of this questionnaire can be found in Appendix R. An article with an invitation to participate has been sent to all organizations and websites that reported the former research of the author about solar panels. An example of the article can be found in Appendix C. Because this strategy did not deliver enough respondents, additionally invitations to join the research were dropped of personally at a thousand dwellings that are suitable for a PV system throughout the Netherlands. An example of the invitation letter can be found in Appendix D. Together this resulted in a total amount of 265 respondents.
With subsidies neglected, (respondents was asked to assume no subsidies are present), only 86 of these respondents were hypothetically willing to invest in a PV system if they had the time and money. 59 respondents did already have a PV system. These two groups together resulted in 146 respondents suitable for the experiment. They were automatically redirected to part two of the questionnaire. The fact that 146 of 265 respondents would invest in PV or had a PV system already has no statistical meaning since the respondents gathering focused explicitly on people interested in PV systems.
In part two These 146 respondents are presented with the nine treatment combinations, i.e. the constructed hypothetical situations. The situations are presented in random order. On top of the 9 choice sets in part two there are 8 direct questions about:
Whether the respondent has a PV system. If not, why not?
If not, whether the respondent would invest in one, when having enough time and money.
The most important motive one would have if one would invest in PV. The year of birth of the respondent.
The mortgage type of the respondent. The education of the respondent.
These seven questions deliver the descriptive statistics of the respondents. 87 people indicated they were willing to invest in a PV system in the present situation. One of the hypothetical scenarios represented the present situation, but 22 respondents indicated that they would not invest in PV in this very situation. 17 of these people chose not to invest in PV in all nine situations. All these 22 respondents are inconsistent and are removed. All 59 respondents in possession of a PV system did also indicate to be willing to invest in PV in the present scenario. This makes the final amount of respondents:
.
There are different ways of checking whether the amount of respondents is enough. Most experiments from the literature (Louviere, 1994; Hensher, 2005; Sormani ,2011; Glumac, 2012) use the rule of thumb that the least chosen options should at least be chosen 30 times. The option, exept for the present situation attribute, that is least chosen is to choose not to invest in PV anymore if all levels are in the middle. Only 20 respondents then opted not to invest in PV. This means that actually 50% more respondents are needed. It was
28
chosen to be very strict in the selection of respondents by checking consistency and their willingness to invest in PV. This strict selection is expected to diminish bias. But on the other hand it led to a shortage of 50% more suitable respondents. However this rule of thumb is a “one-size-fits-all” rule. In this research all scenarios are very much related since they exist of several numerical series of returns on an interval scale. If one series is not supported by enough respondents, the relation with the other series will still provide a sufficiently clear pattern in perceived utility and choice.
4.2 Model fit
Binomial Logit has other rules than ordinary least squares regression (OLS-regression). It is consequently not possible to calculate the adjusted R2. Instead the Mc Fadden rho-square is used (ρ2). This value cannot be compared with R2.
∑
∑ (6)
The rho function deducts a ratio of 1.
The nominator of this ratio is the sum of the natural logs of all chances calculated by the model for the chosen alternative. If for example the chance that a respondent would choose to invest in PV in a particular situation is predicted 0,8, and then the respondent chooses to indeed invest in PV the score is , if not the score is . The scores of all answers are summed up.
The denominator is the sum of the natural logs of all chances that the alternative would indeed be chosen given that the chances would be even over all alternatives. This is .
Values of a very good fit in MNL-regression are between 0,2 and 0,4. (Kemperman,2000; Louviere, 2000).
4.3 Binomial logit model estimation
To create a definite model first it should be checked whether the effects of the returns on stock, deposit and mortgage on utility are linear or non-linear. The aim of the model is not to maximize the fit of the model. This may lead to a model that is very complicated and probably merely fitting the data of this particular sample of respondents. The aim is to understand how other investment alternatives have an effect on utility and thus an effect on choice.
To estimate whether the effects of the independent variables on choice are linear or non-linear a non-non-linear model is made:
29
The coding is displayed in table 4.1 The lowest returns are the omitted levels that are coded minus one.
CODING OF THE ATTRIBUTE LEVELS
Effect coding (1) (2) Linear coding Stock returns 2% -1 -1 -1 6% 1 0 0 10% 0 1 1 Deposit returns 2% -1 -1 -1 4% 1 0 0 6% 0 1 1 Mortgage interest 3% -1 -1 -1 6% 1 0 0 9% 0 1 1
Table 4.1. Coding of the attribute levels, effect coding is needed for the
modeling of non-linear effects with two estimates.
The Nlogit commands and output for the binary logit model above are presented in Appendix E the estimates are displayed in table 4.2.
MODEL ESTIMATES
Variable Coefficient Std. Er. β /std. r. P [|Z|>z]
β0 1,36703810 0,07934528 17,229 0,0000 βstock1i 0,02652824 0,10856460 0,244 0,8070 βstock2i -0,39504178 0,10396722 -3,800 0,0001 βdeposit1i -0,07793625 0,10743945 -0,725 0,4682 βdeposit2i -0,30223764 0,10483677 -2,883 0,0039 βmortgage1i -0,05561766 0,10777816 -0,516 0,6058 βmortgage2i -0,58585457 0,10236503 -5,723 0,0000
Table 4.2. Model estimates for the non-linear model.
It is difficult to interpret the variables of table 4.2 because one has to combine the estimates with the coding. Table 4.3 allows one to read the information from table 4.2 in an easy way with the effects for the omitted variables included. Table 4.3 shows that the effect of changes in the returns on stock, deposits and mortgage payoff is close to linear. The effect on utility of investing in PV when a low return of another investment class increases to a middle return is almost equal to the effect when a middle return of another investment class increases to a high return. The non-linear estimate of the middle level is highly insignificant; the middle level estimate of the deposit attribute is most insignificant of all middle estimates with 47% confidence. This means that the chance to find this or a higher estimate when the real relation is in fact zero is 53%.
The equidistance and insignificance of the middle estimates shows that there is a linear relation between main effects and utility. This does not mean that there is a linear relationship between main effects and choice probability since this relation is given by equation (4). This results in an increasing effect of rising alternative returns on choice.
30
The model fit of the linear and non-linear models are equal; with a model fit of 0,2828 for the linear model and a model fit of 0,2833 for the non-linear model.
The linear model with all interaction effects included has a rho of 0,2984 (Appendix B). The slight increase in fit for this more advanced but collinear model strengthens the assumption that interaction effects are absent.
Table 4.3. The effects of the different other returns on the utilty of
choosing the PV system. The utility of not choosing the PV system is zero.
The linear model is then:
The Nlogit commands and output for the linear model are presented in Appendix F the estimates are displayed in table 4.4.
MODEL ESTIMATES
Variable Coefficient Std. Er. β /std. r. P [|Z|>z]
β0 1,35991535 0,07813996 17,404 0,0000
βstocki -0,36688143 0,09250151 -3,966 0,0001
βdepositi -0,33381051 0,09230260 -3,616 0,0003
βmortgagei -0,59837434 0,09453658 -6,330 0,0000
Table 4.4. Model estimates for the non-linear model.
This is the final model. There are three negative linear relations between a rise in return of an alternative investment and the utility of investing in PV. This makes sense because the higher the return on other investments the smaller the relative utility of investing in PV. The positive sign of the intercept (β0) means that the respondents in the sample on average
PARTH-WORTHS OF ATTRIBUTE LEVELS
Attribute Attribute -level Worth
Choose Choose a PV system + 1,3670 *
Choose no PV system + 0,0000 * Stock returns 2% + 0,3685 * 6% + 0,0265 10% - 0,3950 * Deposit returns 2% + 0,3801 * 4% - 0,0779 6% - 0,3022 * Mortgage interest 3% + 0,6415 * 6% - 0,0556 9% - 0,5858 *
31
attributed more utility to investing in PV than not investing in PV. This makes sense because all respondents in the sample have a PV system or are willing to invest in one. Table 4.5 allows one to read the information from table 4.4 in an easy way. The model has different steps between return levels: 4% for stock, 2% for deposit and 3% for mortgage. Keeping this in mind, the most important conclusion that can be drawn is that the effect of 1% rise in mortgage returns is the highest, second comes the effect of rise in deposit returns and last the effect of rise in stock returns. More about this can be read in chapter 4.5.
Table 4.5. The effects of the different other returns on the utility
Of choosing the PV system. The utility of not choosing the PV system is zero.
What do these estimates mean? When applying the formula:
| (5)
It means that if someone is considering to invest in a PV system and the alternative returns are as they are now: 2% stock return, 2% deposit return and 3% mortgage interest, then the chance that one will invest in PV is 93% and not 100%. This is incorrect of course because all respondents indicated that they would invest in PV. The reason for this is that the model uses the concept of utility. Because everyone chose to invest in PV in this situation the utility of PV in this situation is very high. However we assume a logistic distribution of individual utility, or the error term. Therefore there is a chance that people would still choose not to invest. This mimics true behavior in fact. Since not all people that indicated that the will invest will actually do so in real life. At the other extreme: the situation where all alternative returns are maxed out: 10% stock return, 6% deposit return and 9% mortgage interest, then Only 52% of respondents are still willing to invest in a PV system.
4.4 Descriptive statistics
The social demographics and answers on the direct questions are displayed in table 4.6. The division of the respondent can be particularly interesting because it is possible to run a
PARTH-WORTHS OF ATTRIBUTE LEVELS
Attribute Attribute -level Worth
Choose Choose a PV system + 1,3599 *
Choose no PV system + 0,0000 * Stock returns 2% + 0,3669 * 6% + 0,0000 * 10% - 0,3669 * Deposit returns 2% + 0,3338 * 4% + 0,0000 * 6% - 0,3338 * Mortgage interest 3% + 0,5984 * 6% + 0,0000 * 9% - 0,5984 *
32
separate maximum likelihood estimation for one group if this group is large enough. The different socio- demographic information about the respondents is discussed now. In the order of table 4.6.
Table 4.6. Descriptive statistics
•Possession of a PV system.
As mentioned in 2.3 it is important to have both people that do and people that do not have a PV system in the sample of respondents. The reason for this is that respondents that do not have a PV system can lead to biased results because they say they are interested in buying a PV system, but their stated behavior may not be their true behavior. They could in practice be less interested in investing in PV and thus they could be overly react on changes in alternative returns.
People that do have a PV system may also lead to biased results. Before they invested in PV there were all kind of uncertainties and doubts. Now that the system is operative and embraced it could be the case that these people would miss the system if it would not be
CHARACTERISTICS OF RESPONDENTS % #
Possession of a PV system: -Yes 48% 59
-No 52% 65
>Reason for not having PV: -No money available 26% 17
-Did not get into PV yet 46% 30
-Right now my roof is not suitable 23% 15
-Until now it is not attractive enough 5% 3
>Would get PV if having time and money: -Yes 100% 65
-No 0% 0
Most important motive if one would invest: -Idealism 42 34%
-Diminish risks/less dependency energy prices 21 17%
Good investment 52 42%
Pioneering/ The image 9 7%
Mortgage type: Variable interest, amortization ok. 13% 16
Variable interest, amortization hindered 2% 3
Fixed interest (>2yr), amortization ok. 41% 51
Fixed interest (>2yr), amortization hindered 27% 33
No mortgage 17% 21
Education: Primary school and lower 2% 2
MBO 14% 17
HBO 46% 57
University 38% 47
Other 1% 1
Date of birth: Average 1962
Median 1962
Lowest 1926
33
present anymore. Therefore it is expected that the sensitivity to the increase of alternative returns is less than reality for the respondents that possess a PV system.
The modeling of both separate groups is displayed in table . The models for both groups are again checked for non-linearity. The Nlogit output of these four models can be found in Appendices G, H, I and J. It appears that again that a linear model is most appropriate considering model fit and significance of the β- estimates.
MODEL ESTIMATES
NO PV PV
Variable Coefficient P [|Z|>z] Variable Coefficient P [|Z|>z]
β0 0,91219264 0,0000 β0 2,07027796 0,0000
βstocki -0,31539151 0,0070 βstocki -0,53478551 0,0013
βdepositi -0,39080237 0,0008 βdepositi -0,25399361 0,1139
βmortgagei -0,57932924 0,0000 βmortgagei -0,73134866 0,0000
Table 4.7. The effects of the different alternative returns on the utilty of choosing the PV
system for the respondents that have or have no PV system. The utility of not choosing the PV system is zero.
The results show that the hypothesis that people in possession of a PV system would be more willing to buy such a system again, in a hypothetical situation with higher returns on alternative investments than people without a PV system should be accepted.
People with a PV system have a higher probability to invest again in PV than people without a PV system in any situation. On top of that the proportional change of the forecasted number of PV investors as a result of a change in alternative returns is always higher for respondents without a PV system relative to respondents that do have a PV system.
The crucial question is now which group displays the true PV consumer’s behavior that could be indicated by the development of PV sales in the case of economic change. There are different justifications in favor of both groups. On the one hand it is the case that people that have a PV system have probably become more positive about the system after they bought it. But on the other hand it is very plausible that only a small part of the people that indicated they would be willing to buy a PV system is truly interested in doing so. In this thesis no choice is made between the models because it is too hard to find solid arguments to do so. Therefore the aim is to draw a picture of behavior that is as complete as possible. Models of both groups will be used from now on. Appendix K shows the different probabilities for all 27 possible scenarios for the model of respondents that have a PV system, the model of respondents that do not and the overall model.
Another difference is that the effect per percent increase in deposit returns is slightly higher than the effect per percent increase in mortgage interest for people that do not have a PV system. In the general model and “PV owner” model the order of impact is firstly mortgage interest, secondly deposit return and thirdly stock return.
•If not in the possession of a PV system, the Reason for not having a PV system.
-For 46% of respondents the reason they did not invest in a PV system yet is because they did not get into it yet. This is something heard a lot when ringing people’s doors. People are
34
interested, but they do not trust companies with their advice. What they want is objective information of a neighbor or acquaintance.
-26% of respondents indicated that they did not have the money available to invest. Seeing investing not as an investment but merely as the cheap purchase of energy for the coming 20 years may change people’s opinion regarding budget in the future.
-The roof problem does obstruct 23% of respondents from investing in PV. Reasons could be a sloped roof with lousy orientation or a lot of shadows caused by chimneys for example. But also the “welstandscommissie”, the Dutch guard of esthetics of the built environment, does occasionally forbid deployment of PV. Both reasons often play a role in the case of dwellings built before WWII.
•Whether the respondent would get a PV system if he or she had enough time and money to get it installed. This variable was used to select respondents. Therefore the outcome for the selected sample was consequently 100%.
•The most important motive to invest in a PV system.
Most respondents (42%) indicated that the investment value is the most important motive to invest in PV. 71% of this group are the people that do not have a PV system. These 71% are relatively less willing to invest in PV in general and more sensitive to changes in return compared the respondents that did not have a PV system either but had other motives. This is to be expected. Appendix L displays the model output.
•Mortgage type.
The different mortgage types can be divided in two categories. The types that should be sensitive to changes in the mortgage interest and the ones that should not, at least not on the short term. These groups cannot amortize, or they do not have a mortgage. It appears that the sensitivity for changes in mortgage interest for these groups is less, but still present. Even for the people that do not have a mortgage. Model output is given in Appendix M and N. But the effects of deposit return changes are relatively large. This may indicate that respondents associate changes in mortgage interest with other dwelling market related developments, and that people that cannot amortize or do not have a mortgage attach more importance to deposit returns.
•Education.
The education level of the sample is quite high when compared to the Dutch average. The reason for this is that all respondents own a dwelling and are interested in PV systems. They are in fact relative early adopters. When the estimates are defined only for respondents with a University degree it appears that they are very sensitive to changes in mortgage interest. Appendix O shows the output.
•Age
Lastly it is possible to compare the age of the oldest half of respondents with the youngest half. The youngest half of respondents Is responsible for 44% of the amount of PV possession. The younger people are however much more positive about PV. They attribute much higher utility to investing in PV relative to the older half. The model of the younger half is displayed in Appendix P.
35
In this chapter it was tried to identify patterns in the data that makes us understand the behavior regarding PV investing beyond the mere main model. These patterns provide useful information for the conclusion and reflection of the thesis.
4.5 Cross-elasticities and graphic display of results.
The cross- elasticities can now be calculated for the models. The meaning of the cross elasticity is done a bit different from price-elasticity for example, this is because returns are already a percentage. Here it is the effect of an increase in return of one percent (for example return on stock increases from 4% to 5%) on the amount of PV investors. For example: a value of -0,37% means that an increase in stock return of 1% is expected to lead to an decrease in PV investors of 0,37%.
The effect on utility of increases in alternative returns is modeled linear. The effect on the chance that one will invest in PV is not. This means that the cross- elasticities do not have a fixed value, but they vary with the values of the returns. Because the linear effect on the utility has been proven, the utilities will be interpolated. That means that the points between the levels are constructed for levels that have not been measured such as stock returns of 8% for example.
First the formulas that calculate all cross- elasticities are derived. Next, because this formula is not easy to interpret, the effect of the first percent rise from the present situation is calculated. Lastly the course of the amount of investors in PV as a function of each alternative is sketched keeping the other two alternatives fixed at the present levels.
The deduction of the formula to calculate the cross- elasticities can be found in Appendix Q. Here the elasticity formulas for the PV owner and non-PV owner can also be found.
In this chapter only the three cross- elasticity formulas for the general model are displayed. For stock return:
For deposit return:
For mortgage interest:
With being stock return, being the deposit return and being the mortgage interest. These formulas are rather theoretical. The effect of the first percent rise from the present situation is calculated for each alternative. This is displayed in table 4.8.
36
EFFECT ON PV INVESTMENT OF 1% INCREASE FROM PRESENT RATE
Stock 4%>5% Deposit 2%>3% Mortgage 3%>4%
PV -0,22% -0,36% -0,74%
No PV -0,48% -2,11% -2,08%
Overall -0,37% -1,17% -1,42%
Table 4.8 The effect of 1% change in alternative returns.
The course of the amount of investors in PV as a function of each alternative keeping the other two alternatives fixed at the present levels has been sketched in figure 4.9. Modeled probabilities for the lowest levels are divided by a factor to equal 100%. The most important outcome is that the mortgage interest has the largest effect, but that still the investment in PV is relative insensitive to the alternative returns: If the present mortgage interest would double PV investment decreases only with 3% to 7%. If more alternatives increase together there are however synergy effects. All effects together are stronger that the sum of the separate effects. Also the difference between PV owners and people that do not own PV is striking. This relative difference has the largest for deposit returns.
37
Figure 4.9. The graphic display of the share of present investment as a function of the return
of one alternative keeping other two alternative at present levels. 80% 85% 90% 95% 100% 105% 0% 2% 4% 6% 8% 10% 12% Sh ar e of p re sent P V in ve st m e n t Return on stock PV owner Overall Owns no PV 80% 85% 90% 95% 100% 105% 0% 2% 4% 6% 8% 10% 12% Sh ar e of p re sent P V in ve st m e n t Return on deposit PV owner Overall Owns no PV 80% 85% 90% 95% 100% 105% 0% 2% 4% 6% 8% 10% 12% Sh ar e of p re sent P V in ve st m e n t Mortgage interest PV owner Overall No PV owner
39
5. Conclusions
5.1 Introduction
There is a transfer going on to a more sustainable footprint of our society. Or at least a lot of people and companies are engaged in this transfer. The drivers that underlie are sometimes a bit unclear. Is it idealism or hard currency? Or a combination of both?
The field of energy consumption and energy production has an important place in sustainability. In the Netherlands energy generation of photovoltaic panels or PV panels grows the fastest of all energy generation alternatives. The conditions for PV investing are very attractive at the moment, since there are not much investment opportunities for retail investors due to the economic crisis. This caused dwelling owners to invest over €200.000.000 in PV in 2012.
In the survey conducted 42% of all respondents indicated that fact that PV is “a good investment” is their most important motive to invest. How will investments in PV develop if other alternatives become more attractive?
The research has established a first exploration of the behavior regarding PV investing. The amount of energy generated with PV panels grew with 64% over 2011 and 138% over 2012. In 2011 60% of these panels were installed on owner occupied dwellings. That does not only mean that it is inevitable that the presence of a PV system will become an increasingly common dwelling-attribute in the nearby future, but also an increasingly common investment vehicle available to retail investors. So far it was only researched what the considerations are of households regarding PV systems as a consumer product, not as an investment vehicle.
As expected a priori there is a difference in the behavior of people that already have a PV system and the people that do not. People in possession of a PV system are less sensitive to the increase in the returns of the alternatives stock, deposits and mortgage amortization. Because it is hard to tell which group represents the true amount of newly installed PV systems best, for both groups and the overall group a model is made.
Both groups have arguments in favor of themselves:
•The people that do have a PV system already embraced the new technology. The possible risks of investing are conquered mostly. This may cause investors to be more positive about their PV system than they would be a priori. True market behavior is probably a bit less positive regarding PV. However when these people relocate it is very well possible that they would re-invest in PV, also driving up PV investment.
•The people that indicated that they have no PV system but would invest in a PV system if they had the time and money free to invest might not do so in reality. It is possible that only a part of these respondents is a real “PV investor”. The ones that are not that serious are of course much more sensitive to the rise of alternative returns. This may have cause a sensitivity to alternative returns that is higher than true market behavior.
About respondents that do have a PV system this doubt does not exist. They are real PV investors a 100% sure.
40 5.2 The model
The most important outcome is that the investment in PV systems by dwelling owners is not very sensitive to the increase of one alternative return, but that there is indeed a significant relation.
Sensitivity is a bit lower for respondents that indicated to have idealistic motives but even respondents that indicated that their most important motive was financial were not very sensitive to the changes in alternative returns. This suggests that all PV investors perceive some “non-monetary return” of investing in PV.
PV investment turned out to be most sensitive to increases in mortgage interest. But the effect is not very large: if the variable mortgage interest would rise from 3% to 6% more than 90% of all respondents (having PV, not having PV and overall) would still invest in a PV system. If however all alternatives’ returns are maxed out only between 41% and 63% still invest in PV depending on which model is used. The overall model has a probability for PV investing in this scenario of 53%.
It is quite surprising that there is a synergy-effect: the effect of all returns increasing is larger than the sum of all separate increases. The maxed out situation was actually put in the design and presented to all respondents. This is a good opportunity to check the model. 58% of all respondents would still invested in PV. This shows that the theoretical model predicts the true behavior of respondents quite well.
The effects of alternative returns are ranked in table 5.1.
EFFECT ON PV INVESTMENT PER % INCREASE OF RETURN
Stock Deposit Mortgage
PV 3rd 2nd 1st
No PV 3rd 1st 2nd
Overall 3rd 2nd 1st
Table 5.1. Ranking of the magnitude of effects of alternative
Returns on PV investment.
Mortgage has the highest impact overall and for PV owners. This makes sense since most respondents (83%) have a mortgage, and mortgage interest is likely to be quite a large expense. For people that do not have a PV system deposit returns have the largest effect, but only slightly higher than the effect of mortgage interest. This may have to do with the fact that people that do not have a PV system still associate investing in PV with risk. They might think that if deposit returns are high they can put money on the deposit account and wait for PV technology to become cheaper and have less start-up problems. Or maybe they wait for new subsidies to be announced. It is easy to withdraw the money and invest in PV after all.
5.3 Further findings
Apart from the relative low impact of single changes and the large impact of combined changes most outcomes of the research were as expected. The difference in attitude towards PV systems between younger and older people was remarkable. Earlier research
