Renewable and Sustainable Energy Reviews Regulation, pro ﬁ tability and di ﬀ usion of photovoltaic grid-connected systems: A comparative analysis of Germany and Spain

Although both Germany and Spain were pioneers in the support and adoption of photovoltaics (PV) through Feed-In Tari ﬀ s (FiTs), Germany is the world's leading country in terms of installed PV capacity per capita and remains on a path of PV growth while the Spanish PV market is virtually paralysed. We analyse the divergent performance in both countries by reviewing the evolution of FiTs and by analysing the pro ﬁ tability of di ﬀ erent types of installations for the period 2004 – 2014. While in Germany the policies, pro ﬁ tability and di ﬀ usion of PV were relatively stable over time, in Spain, erratic policy changes and the distortionary e ﬀ ect of the solar orchard ownership structure caused a di ﬀ usion bubble (June 2007-September 2008) and a subsequent pro ﬁ tability bubble (2009 – 2011). These bubbles led to the collapse of the support system in 2012, revealing the di ﬃ culty inherent in designing innovation and di ﬀ usion policies in the context of high uncertainty and rapid technological change.


Introduction
During the last decade, photovoltaic (PV) technology investment has taken off thanks to its rapid cost decline ( [45]:23; [108]:10) as well as policies carried out across the globe to boost its diffusion [43,89]. Although in 2015 photovoltaics accounted for only 1.3% of the global electricity generation [44], this share is expected to increase in the context of the decarbonisation of energy production. Specifically, the International Energy Agency (IEA) estimates that the share of PV in global installed capacity will reach 13% by 2040 under the new policies scenario, or even higher if countries aim to meet the 2°C target set in the Paris agreement [41,42]. PV has the highest technical and sustainable potential (i.e. taking into account ecological constraints) amongst all renewable energy technologies ( [69]:1159; [107]:119), and is considered by many authors as able to serve as the main energy source in a decarbonized energy system [5,70,93]. Indeed, if appropriate policies are implemented, electricity from photovoltaics could cover up to 90% of total global energy demand by 2050 (e.g. [35]).
Germany has the highest installed PV capacity per capita (511 W/ capita according to IEA [40]) and ranks second in absolute value after China (43.5 and 39.7 GW respectively [41]). Spain, which accounted for 45% of the global newly installed capacity in the year 2008 (see annual installed capacities in Fig. 2), has had a virtually paralysed PV market [102][103][104][105]2,3] until the renewable tenders of 2017. While the determinants of PV diffusion in Germany have been identified and quantified in the literature [24,90,91], this type of econometric analysis is not possible for Spain due to the diffusion and profitability bubbles and the recurrent and sudden policy changes. Thus we compare the cases of Germany and Spain in terms of regulation and the profitability provided to PV investors through Feed-In Tariffs (FiTs), and how the evolution of the expected profitability of PV was accompanied by the diffusion of PV systems, in order to disentangle the determinants of the divergent evolution of PV in both markets. We study in detail the impact of the "solar orchard" ownership structure in Spain (large installations divided into small ownership units in order to minimize costs while still obtaining the higher FiT for small installations), in order to understand the investment boom that occurred between June 2007 and September 2008 and the subsequent collapse of the support system. We investigate not only how this ownership structure boosted profitability, but also how it created at least 374 M€ of extra policy cost 1 in 2008 alone.
We estimate the expected profitability through the internal rate of https://doi.org/10.1016/j.rser.2018.04.030 return (IRR) for three different types of systems (small, medium and large) for the period 2004-2014. We illustrate the potential for excess profits by comparing the estimated IRR to the required cost of equity (COE) estimated through the capital asset pricing model (CAPM). The paper is structured as follows: Section 2 briefly reviews the evolution of the regulation and PV support mechanisms in both countries. Section 3 presents the methodology and data for the calculation of the internal rate of return and of the cost of equity, differentiating between static (assumed constant) and dynamic determinants of profitability. Section 4 presents the results of expected profitability and relates them to the diffusion of PV technology in both countries. For the case of Spain, the role of "solar orchards" is closely studied. Section 5 provides a sensitivity analysis for the two dynamic determinants of profitability (system and electricity prices) to see how PV profitability could evolve in the future depending on the evolution of these parameters. Finally Section 6 presents the conclusions.

PV support regulations in Germany and Spain
Feed-in Tariffs (FiTs) have been the most important instrument for the promotion of photovoltaics and other renewable energy technologies in Europe [36,6]. While Feed-in Tariffs or Feed-in Premiums may have different structures [12,77,92], the basic idea is that a preferential price is paid to renewable energy producers in order to ensure a certain level of profitability. This in turn triggers private investments which would not have happened otherwise. Although FiTs have generally been an effective instrument for the promotion of renewable technologies [37], flawed design may lead to the collapse of the system [87]. Although FiTs have been the most important instrument, other mechanisms such as tax incentives and tradable green certificates [1], and investment subsidies and soft loans [83] have also been implemented. These incentive mechanisms were not generally implemented in isolation, but combined to achieve more effective results [22,88].
While there is no optimal policy for the promotion of renewable energy technologies, we can conclude from the comprehensive studies of Avril et al. [4] and Ming-Zhi Gao et al. [82] that there are three main steps in the development of a consistent renewable energy policy: (i) a first phase focusing on R&D and demonstration projects; (ii) a second phase implementing FiTs and other instruments to promote market penetration, and (iii) a final move towards more competitive incentives such as tendering schemes (see [33] for a comparison between FiTs and tendering schemes) as the technology becomes mature. It is relevant to stress the importance of adapting the remuneration levels in the second phase to the cost evolution of the technology, for example, through the use of responsive FiTs to avoid unnecessary policy costs [34]. Since the historical evolution of PV support policies in Europe [25,75] and the world [31,39,74] has been widely studied in the literature, we focus on the cases of Germany and Spain. These two cases are also studied in terms of the pdynamics of collective expectations about PV technology by Kriechbaum et al. [74].
In Germany, the climate and energy policy has been developed through the so-called Energiewende (Erneuerbare-Energien-Gesetz or EEG) [71]. Despite the numerous reforms of the EEG, the policy has been consistent (based on FiTs with a degression rate 2 ; see Fig. 1 for a simplified illustration or Fig. A1 in the Appendix A for a detailed structure and evolution of the FiT levels in Germany) and stable compared to the abrupt and even retroactive policy changes that occurred in Spain.
Hoppmann et al. [38] identify four phases in the evolution of the German FiTs for PV: (i) a first phase until 2000 based on R&D, local initiatives and two large demonstrations programs; (ii) a second phase between 2000 and 2004 when FiTs became the main instrument for the promotion of PV, but with still modest installed capacity due to high costs; (iii) a third phase between 2004 and 2011 when installed capacity grew substantially and the government tried to limit the rising costs for society; and (iv) a fourth phase from 2011 onwards when the government further reduced incentives to control the market penetration of PV. As of 2017, only rooftop systems are still eligible for FiT. Groundmounted systems below 10 MW can participate in tenders which, as of the time of this writing (October 2017), have been granted with a mean remuneration of 4.91 €c/kWh and a record low of 4.29 €c/kWh. Ground-mounted systems above 10 MW do no longer enjoy financial incentives.
While the four phases identified by Hoppmann et al. [38] for Germany were consistently based on a FiT system with degression, we can identify 3 phases in Spain with more disruptive policy changes: (i) The first phase until September 2008 was characterized by FiTs, in addition to other instruments such as soft loans (with interest rates around 3%) and investment subsidies (up to 40%) until 2005 [16,18,21]. In the last months of this period (June 2007 -September 2008) Spain experienced an investment boom when around half of the current installed capacity was deployed [19]. (ii) To control the further diffusion of PV, the government changed the category definitions from capacity (≤100 kW, 100 kW to 10 MW, and 10-50 MW) to capacity and type of installation (roof <20 kW, roof 20 kW to 2 MW, and ground-mounted <10 MW) and implemented a system of (lower) FiTs with degression and capacity caps, which was active between October 2008 and December 2011 [17,23]. (iii) Finally, a Moratorium was passed in January 2012 such that all incentives for PV were removed. Additionally, cost-containment mechanisms were implemented between 2010 and 2013, undermining the profitability of already existing installations [83,99] and increasing the investment risk for future potential projects by increasing legal uncertainty [104,105]. Although the PV market has been virtually paralysed in Spain since the 2012 Renewables moratorium, there have been two renewable auctions in 2017 for a total of 8037 MW capacity, of which 3910 MW have been allocated to PV.
These developments in both Germany and Spain confirm the tendency suggested by Ming-Zhi Gao et al. [82] towards tendering schemes rather than FiTs to support the further diffusion of PV. Tendering systems are becoming predominant over FiTs, with at least 67 countries implementing some type of renewable auction ([67]:32). There are three main differences between the German and Spanish tendering systems: Although most design elements involve trade-offs between policy objectives (e.g. neutral auctions favor cost efficiency but can create imbalances between technologies, as in the first Spanish auction in 2017 when PV only obtained 1 MW of the 3000 MW auctioned), [20:11] identifies some "best practices" such as "schedul[ing] of auctions, volume disclosure, price ceilings, penalties, streamlin[ing] of administrative procedures and provision of information to potential participants".
Other alternatives to tenders are (i) power purchase agreements (PPA), through which two parties (generator and consumer) agree on a long term electricity supply contract, and (ii) lease contracts, through which the consumer pays a pre-determined monthly fee to the owner of the system independently of its generation. These market-driven financing schemes are more common in the USA (particularly California, see e.g. Davidson et al. [14]) than in Europe but are likely to also become more important in the latter region as the technology reaches grid parity.
3. Method and data for the estimation of the expected profitability and cost of equity Nofuentes et al. [85] study how different profitability indicators, such as the net present value, the (discounted) payback time or the profitability index can be applied to PV. Talavera et al. [100] carry out a comprehensive sensitivity analysis on the determinants of PV profitability for the Euro area, Japan and the USA. Their results are interesting for this work, as they show how different factors affect the profitability of PV depending on different market and regulatory conditions. In the three cases, they find that the initial investment and the PV yield/electricity price (both variables have the same quantitative effect on the IRR) are the most important factors determining the profitability of PV investments. Initial investment (system price) and electricity prices (FiTs) are indeed the dynamic determinants of profitability on which we will focus in this paper.
The most recent literature on the profitability of PV grid-connected systems in Germany and Spain is that of Talavera et al. [99] for the case of Spain, Spertino et al. [96] for Germany and Italy, and De Boeck et al. [15] for the major European markets (Flanders (Belgium), Germany, Italy, Spain and France). Talavera et al. study the cost and profitability evolution of PV in Spain for the period 1998-2013, using the internal rate of return (IRR), the net present value (NPV) and the levelized cost of electricity (LCOE). We review and expand this analysis by including the effect of solar orchard ownership structure on the profitability of PV installations, which explains why large systems were deployed despite having apparently negligible profitability ( [99]:242). Spertino et al. estimate the profitability (both IRR and NPV) for rooftop systems in Germany and Italy between 2006 and 2013. De Boeck et al. [15] also study profitability of residential systems, contrasting the stable evolution of the German PV market against the volatility of the Spanish case, and concluding that the latter is the only one where residential PV systems are not currently profitable. Finally, López Prol and Steininger [81] assess the social profitability of PV in Gernany by including integration costs and environmental externalities, concluding that PV is socially profitable and therefore incentives are economically justified.
Other analyses focus on the profitability of PV for self-consumption applications, such as Colmenar-Santos et al. [11], Chiaroni et al., [7], Talavera et al. [98] and López-Prol and Steininger [80], Finally, Cucchiella et al. [13] and Lang et al. [77] focus on the profitability of PV systems for self-consumption without subsidies, suggesting that PV selfconsumption is already profitable for a wide range of segments and applications; i.e. socket parity has been achieved.
Below, we present the method of our analysis, starting with the definition of categories, specifying the calculation of the IRR, its determinants and the estimation of the cost of equity (COE) through the capital asset pricing model (CAPM).

Defining categories
Although perfectly consistent categories of installations cannot be defined due to the differing structures and evolution of the FiT systems in both countries, we can define comparable categories by assigning homogeneous criteria to the different types of installations: small-, medium-and large-scale installations. Therefore, we assign residential system prices (installation cost) for small installations, commercial/ industrial prices for medium installations and utility-scale system prices for large installations, all as provided by the IEA. 3 When these categories are not explicitly mentioned, but a range is provided, we take the upper bound of the range as the system price for small installations, the lower bound for large installations and the average for medium installations.
Likewise, we assign the highest FiTs to small, the lowest to large and the intermediate FiT level to medium installations. Finally, the financing costs are the 2004-2014 average interest rates for household purchase (small scale), loans below 1 M€ (medium scale) and loans above 1 M€ (large scale) for each country, provided by the European Central Bank. Although these segments are somehow representative of residential, commercial-industrial and utility segments respectively, we avoid naming them as such because, due to the aforementioned category changes, these categories do not exactly coincide with those segments. A summary of the input data can be seen in Fig. 1 and Table 1.

The internal rate of return (IRR)
Although there are different indicators of profitability [85], we employ the internal rate of return because (i) it is relative and therefore easily comparable with any other type of investment, regardless of investment amount or lifespan, (ii) it does not force us to make arbitrary assumptions about the discount rate, and (iii) it is the most widely used indicator by investors and policy makers. While the project IRR does not consider financing costs, the equity IRR includes the cost of the external capital, therefore indicating the profitability for the final investor. The criterion for carrying out the investment is that the project IRR is higher than the weighted average cost of capital, or that the equity IRR is higher than the cost of equity. We include financing costs in our calculations, and thus we estimate the expected profitability for the equity investor, which we then compare with the modelled cost of equity. The IRR is derived from the net present value (NPV) Eq. (1), which is the sum of the present value of the cash inflows during the lifetime of the investment (PW CIF N [ ( )]) minus the life cycle cost of the investment from the user's standpoint 4 (LCCusp).
The present value of the cash inflows ( PW CIF N [ ( )])is determined by the electricity priceor corresponding FiT -(Pu [€/kWh]), the annual electricity yield of the system (Epv [kWh/kWp y −1 ]), and the discount rate (see below).
In specifying (PW CIF N [ ( )]) in Eq. (2), we also consider the gridaccess charge (γ) and the generation tax (λ) for the case of Spain, applicable since 2011 and 2013 respectively. The last element of Eq. (2) represents the effect of the annual escalation rate of electricity prices (ε PU [%]), the annual degradation rate of the system (dg [%]) and the 3 Data from the annual PV trends reports and the respective national reports [43,44,. 4 The LCC can be considered either from the point of view of the system (excluding investment subsidies, soft loans or any other incentive), or from the point of view of the user including all available incentives [85]:556). As we are interested in the investor's perspective we will analyse the LCC from the user's standpoint. discount rate (d [%]): N being the lifetime of the investment and K PU the discount factor of electricity revenues: On the other hand, the LCCusp comprises the present value of the initial investment of the PV system (PW PV [ ] IN [€/kWp]) and the present value of the annual operation and maintenance (O&M) costs The main parameters regarding the initial investment are the PV system price (PV IN [€/Wp]) and the financing conditions: the percentage externally financed (α), the interest rate (i), and the maturity of the loan (Nl): q being the discount factor: Finally, the present value of the O&M costs is: ε OM being the escalation rate of O&M costs and K PV its corresponding discount factor: In conclusion, we calculate the discount rate (d) of Eq. (9) when = NPV 0: While FiTs are provided for 25 years in Spain (the assumed lifetime of the system), in Germany they are provided for only 20 years. Thus, for the analysis of Germany we need to add the revenues of these last five years at the corresponding wholesale electricity price from year 20, instead of the tariff value. For the sake of simplicity, A now represents the discount factor of the revenues (corresponding to the last element of Eq. (2)) and A' represents the same discount factor for the last 5 years (being now n = 20 and n' = 5). We then have: Where the first addend represents the cash inflows of the first 20 years and the second represents the cash inflows of the remaining 5 years. ′ Pu is the annual average wholesale electricity price (at which the electricity must be sold once the FiTs expire) in the year 20, that is: Finally, to obtain the real internal rate of return (d r ) we simply deflate the nominal IRR (d) by the expected long term inflation rate (π ) e , assumed to be 2% in accordance with the ECB long term inflation target: 3.3. The cost of equity and the potential for excess profits Since we include financing costs in the IRR calculation, the criterion to carry out the investment is that the IRR is higher than the cost of equity (COE). The potential for excess profits 5 would be given by the difference between the rate of return provided by the FiTs and the required COE. We estimate the cost of equity according to the capital asset pricing model (CAPM) [79,95], such that: Where r f is the risk-free interest rate (assumed to be that of 10-year government bond yields), β is a risk indicator which measures the volatility of the returns of the specific security in relation to the market as a whole (1.3 for Germany and 1.54 for Spain according to Noothout et al. [86]) and MRP is the market risk premium (median of 5% and 6% for Germany and Spain respectively, according to Fernandez et al. [27]).

Determinants of profitability
We differentiate between dynamic and static determinants of profitability. Although in reality all of the determinants fluctuate over time, we assume that some of them remain constant (static) to focus on the effect of the dynamic determinants (system prices and FiTs/electricity prices). Table 1 presents a summary of the input parameters for the calculation of the internal rate of return (IRR) along with their sources.

Dynamic determinants
The two main determinants of the evolution of the profitability of PV systems are the electricity prices at which the electricity generated is sold to the grid (FiTs), and the PV system prices (installation costs). The evolution of these parameters for the different categories is shown in Fig. 1. System prices were higher in Spain than in Germany until 2009, when they began to converge. There was a peak in 2007-2008 in Spain due to the investment boom and the consequent peak demand, and a subsequent drop in 2009.
Regarding the Feed-in Tariffs, we can see that the evolution was more stable in Germany, consistent with the evolution of system prices. In Spain, however, the evolution of FiTs was more erratic, with only two categories until May 2007. Between June 2007 and September 2008 (when the investment boom occurred) there were 3 categories (≤100 kW, 100 kW to 10 MW, and 10-50 MW), which were changed in 2009 (roof <20 kW, roof 20 kW to 2 MW, and ground-mounted 2-10 MW), and maintained until 2012. In 2012 the moratorium was passed, entailing that new installations would have to sell the electricity generated to the grid at the wholesale electricity price.   [2:67] suggests that peak power is between 10% and 20% higher than nominal power in Spain, we apply a factor of 1.15 to obtain comparable data. Categories for installed capacity in Spain show the total size of the system, regardless of the ownership units (see Appendix A for raw data according to ownership structure). SO stands for solar orchard ownership structure (Section 4.2). Source for annual installed capacities: [10] and [28].

Static determinants
To isolate the effect of FiTs and system prices on the evolution of PV profitability, we assume that all the other determinants remain constant. Regarding the PV yield (kWh/kWp y −1 ), we assume optimally mounted systems with 75% performance rate at average radiation conditions, with data provided by the PVGIS project [97]. We also estimate the profitability for the extreme values (min. and max. solar radiation) for the best year and installation category to see the range of plausible results depending on the location of the system. While in Germany the electricity generated per Watt peak (Wp) of installed capacity 6 was smoothly increasing since 2000, in Spain there was a leap in 2008-2009 when the electricity yield per Wp more than doubled, probably due to (i) the shift from small to large scale (and therefore more efficient) installations and (ii) the fact that most of installed capacity was deployed in the highest insolation areas (see Fig. A.4 and A.5 in the Appendix A for detailed illustrations on regional installed capacity during 2007-2008 and generation per unit of capacity respectively). The annual degradation of the PV system is 0.8% according to Jordan and Kurtz [72].
Regarding financing conditions, we assume that 80% investment is externally financed at 10 years maturity as indicated by UNEF (private communication) and Fraunhofer ISE [30]. We assume the interest rates to be the average 2004-2014 interest rates for household purchase for small installations, those for loans below 1 M€ for medium-scale installations and interest rates of loans above 1 M€ for large installations, for each country provided by the European Central Bank (see Table 1).
Finally, we assume Operation and Maintenance (O&M) costs to be an annual 1% of the system price with an annual 1% escalation rate [100]. The escalation rate of electricity prices is assumed to be 2% [26].

Evolution in Germany and Spain, 2004-2014
We have estimated the expected profitability at the moment of commissioning (which is not necessarily equal to the realised profitability, particularly for the case of Spain where retroactive cutbacks were implemented between 2010 and 2013 [83,99]). The results are shown in real terms (2% assumed long-term expected inflation rate), after financing costs and before income taxes, to make them as comparable as possible. The system is assumed to be optimally mounted, with average radiation conditions (the ranges between min. and max. radiation levels are shown for the best year and category). Fig. 2 shows that the evolution of both profitability and diffusion of PV systems has been more stable in Germany than in Spain, in consonance with the smoother evolution of FiTs and system prices (Fig. 1). Whereas in Germany the expected real rate of return has mainly ranged between 5% and 12% (except 2013), expected profitability in Spain during the same period has ranged from negative values up to 24% for average radiation conditions. The common feature of both countries is that while smaller systems enjoyed slightly higher profitability in the first years, therefore accounting for a larger share of the total installed capacity, large installations have had higher expected profitability and diffusion rates since 2008-2009.
The results for Germany suggest a correlation between profitability and diffusion, the correlation being stronger for the larger installations. While small installations (mainly residential) have diffused even when profitability was below the cost of equity (COE), large installations have mainly been deployed in the years when expected profitability was higher than the COE (2009-2012).
The case of Spain is more complicated. There were only two FiT levels until May 2007 (this is why the IRR figure shows two red lines up to that year, representing medium installations receiving the high or low FiT levels). Until 2008, the profitability for all categories was below 6.5%. The investment boom happened between June 2007 and September 2008 (represented in the IRR figure in the year 2008) when the intended profitability provided by the FiTs was between zero for large installations and 6.4% for medium installations. However, actual profitability was boosted up to 12% thanks to the effect of the solar orchard (SO) ownership structure, which we will study in detail in the next section. Although FiTs were reduced in 2009, profitability rocketed to almost 24% at average conditions, almost double the COE, due to the even stronger decline of system prices (see Fig. 1). Diffusion did not increase at the same rate due to the cap on installations eligible to receive FiTs set by the government to limit the further expansion of PV and therefore policy costs. In 2012 the FiTs were removed and new installations were required to sell the electricity at wholesale electricity price. Profitability then became negative and new installations virtually zero (most of the installed capacity shown in Fig. 2 corresponds to systems with FiTs allocated in the previous years but installed after 2012).
We can see that the Spanish "profitability bubble" (2009-2011) was subsequent to the "diffusion bubble" (June 2007-September 2008). How is it possible that installed capacity of large installations soared between June 2007 and September 2008 when their expected profitability was virtually zero? Thanks to the effect of the solar orchard ownership structure, profitability expectations rocketed from zero up to 12%. Although several authors have identified this ownership structure as a relevant explanatory factor of the investment boom [23,99], no study to our knowledge has previously quantified its effect on expected profitability and (extra) policy cost (see next section). Indeed, although Fig. 2 shows that the vast majority of the systems deployed in Spain were above 500 kW of installed capacity, they were usually divided into small allotments (≤100 kW) in order to obtain the high FiTs for small installations (see Fig. A.3 in the Appendix A for the detailed diffusion data according to ownership size). Several other factors contributed to the diffusion bubble, according to [19:12], such as (i) political (anticipated policy change to less favourable conditions, inflexible rates); (ii) technological (modular technology, increasing efficiencies); (iii) financial (easy access to credit, favourable €/$ exchange, capital flight from housing) and (iv) administrative factors (quick permissions, poor coordination between different authorities).
These results are consistent with Spertino et al. [96] and Talavera et al. [99] for the cases of Germany and Spain respectively. We obtain higher profitability levels in the peak years, which might be caused by the following factors: (i) We include financing costs, so proftiability is boosted when the internal rate of return is higher than the interest rate on the external financing, and (ii) we estimate the IRR before income taxes to make it more comparable across various types of installations (owned by different agents with different tax rates) and countries.

Solar orchards and extra policy cost in Spain
A "solar orchard" (SO) is a large photovoltaic installation whose property is divided across many investors. Thus, solar orchards can achieve the lower costs of large installations, but obtain the higher FiTs of small ones, boosting profitability above the levels intended by the regulation. We define "extra policy cost" as the difference between the high FiT received by solar orchards and the lower FiT those installations would have received if they were considered to be one large installation instead of many small ones.
A total of 3.7 GWp (or 3.2 GW before applying the 1.15 conversion factor) were installed during the years 2007-2008, half of which were deployed in the three most southwestern regions (Castilla-la Mancha, Andalucía and Extremadura), and 73% in the southern half of the countryincluding islands (see Where 2404 GWh is the electricity generated in 2008 coming from new installed capacity in 2007-2008, S i is the share of installed capacity per type of SO (i), and E i is the extra policy cost per type of SO (i) and per GWh. Assuming that all the SO have the same average efficiency (this is a conservative assumption, since it is likely that larger installations are better mounted and located, and therefore generate more electricity per Wp), the total extra policy cost in 2008 would be 373 M€, that is, 37.7% of the total policy cost and 32.4% of the total revenues received by PV producers in that year, according to the data provided by the CNMC [10]. Thanks to the SO ownership structure, profitability, which was negligible for large installations, could achieve almost 12% in average conditions and a maximum of 16% in the best locations. Only with the effect on profitability of the solar orchard ownership structure can we explain the investment boom which happened between June 2007 and September 2008 in Spain.

Sensitivity analysis and the future of PV
Finally, we carry out a sensitivity analysis on the dynamic determinants of profitability, which allows us to explore the potential future evolution of PV profitability. The base case corresponds to large installations in 2014. We analyse a wide range of possible cases: system prices between 0.3 and 1 €/kWp at 4 and 8 €c/kWh electricity prices on one hand, and electricity prices between 3 and 10 €c/kWh at 0.5 and 1 €/kWp system prices on the other (Fig. 4).
Obviously, Spain has higher profitability than Germany in equal conditions given its higher solar radiation. However, that is partially compensated by the lower cost of equity in Germany than in Spain. At a 4 €c/kWh electricity price, system prices should drop to~0.45 and 0.65 €/Wp in Germany and Spain respectively to achieve a 5% rate of return; or alternatively, electricity prices should increase to~8.5 and 6 €c/kWh respectively at a 1 €/kWp installation cost. While system prices are expected to keep declining [29,45,68], there is more uncertainty regarding the future evolution of electricity prices. Whereas on one hand the merit order effect caused by higher renewables penetration [32,8,9,93,101] puts downward pressure on wholesale prices, potential fossil fuel or carbon allowance scarcities could boost electricity prices.
The evolution of PV system prices in the last years, together with the results of the latest renewable auctions in Germany and Spain (with the lowest price of 4.29 €c/kWh in Germany and a floor price of 3.2 €c/ kWh in Spain as of October 2017), confirm that PV has roughly achieved grid parity in both countries. However, auctions are still necessary in order to ensure floor prices (to protect against the meritorder effect) and to provide higher legal certainty or, as Karneyeva and Wüstenhagen [73] explain, to alleviate increased levels of policy and revenue risk.

Conclusions
We have estimated the expected real profitability (through the internal rate of return) of different types of PV systems at the moment of commissioning in Germany and Spain for the period 2004-2014. Both countries used Feed-in Tariffs as the main incentive to provide positive returns to potential investors, therefore triggering the diffusion of the technology. The results suggest a correlation between profitability and diffusion in Germany, stronger for larger systems, which were mainly deployed when average expected profitability was higher than the cost of equity (2009)(2010)(2011)(2012). In Spain, however, the diffusion of PV was concentrated in a few months (June 2007-September 2008) when roughly half of the current installed capacity was deployed.
We have been able to disentangle one of the main determinants of the Spanish PV investment bubble and the subsequent collapse of the support system by considering the effect of the solar orchard ownership structure (large installations divided into small ownership units in order to minimize costs and obtain the higher FiT for small installations), which accounted for 91% of the installed capacity in 2007-2008. Thanks to that ownership structure, investors could boost profitability from between 0% and 6% depending on installation type, up to 12% at average radiation conditions, causing an extra policy cost of some 373 M€ in 2008 alone (i.e. 32.4% of the total revenues of PV producers in that year). Paradoxically, when FiTs declined between 2009 and 2012, profitability rocketed due to the even stronger decline of system prices. The spike in profitability (up to 24% at average conditions) was not accompanied, however, by a consequent increase in diffusion due to the cap on the installed capacity eligible for FiTs applicable in that period in Spain. The FiT system was finally removed in 2012, which together with the retroactive cutbacks and the corresponding legal uncertainty, virtually paralyzed the Spanish PV market until the renewable auctions of 2017, where 3700 MW were allocated to PV.
This comparative analysis on the regulation, profitability and diffusion of photovoltaic grid-connected systems show the difficulty of designing innovation and diffusion policies in the context of high uncertainty (e.g. response of investors to the FiTs, total policy cost) and rapid technological change (i.e. declining PV costs and increasing efficiencies). While Germany was better able to adapt the support level to the evolution of the PV cost, in Spain the distortionary effect of the solar orchard ownership structure, together with the abrupt policy changes, caused a diffusion and a subsequent profitability bubble that finally led to the collapse of the support system.
Further research should study how the merit order effect caused by higher PV penetration can undermine investors' profitability in the marginal wholesale electricity market once grid parity has been achieved, and more broadly, how electricity market design can adapt to higher shares of zero marginal cost technologies. Besides, batteries and other forms of energy storage could have a disruptive effect on electricity markets and investors' profitability (particularly for PV selfconsumption applications), which would be another interesting research direction.    Installed capacity in Spain has been converted from nominal to peak using a 1.15 conversion factor according to ASIF [2]. Source: own calculations based on CNMC and Fraunhofer data.