An Alternative Approach to Frequency of Patent Technology Codes: The Case of Renewable Energy Generation

3.1 TTs’ Identification Methodology

If equation (1) is strictly positive, then we have a technological state (or technological change) change between periods s and t .

While a distance only allows for pairwise comparisons for a particular order of periods ( r , s , t ) , if we have d ( μ r , μ s ) > d ( μ s , μ t ) > 0 , that indicates a greater technological change between r and s as compared to s and t . If this technological change is significant compared to other periods, we identify it as a potential TT.

TVD is invariant to the number of observations at each period since it depends on the relative distribution. However, the number of observations in a distribution affects its variability. To avoid misguided interpretations related to a change of the TVD that may respond to the total number of observations, we consider two types of periods:

1. Natural periods, in which each distribution includes the observations of one or several calendar years.

2. Periods with the same number of observations (PSNO), in which we partition the data into consecutive subsets, each with the same number of observations, making a further breakdown by month and day.

Note that TVD is not sensitive to category reordering. TVD may vary to a certain degree in the absence of a TT; such variation depends on the nature of the technological field and moment, as well as the total number of observations in each period. Therefore, in absolute terms, we cannot claim any particular TVD level to be indicative of a TT. Instead, we focus on local maxima specific to the technological field under study, which show a greater distance both directions in time.

The application of the TVD as an identification strategy of TTs requires the adequate identification of technological fields, which contain emergent technological innovations. A narrow field selection could exclude potential emergent technological innovations. In our example, the sources that REG use as primary (e.g., wind, water, or sun) have not changed in at least the last century, and we can reasonably argue that those categories are fixed in our period of analysis. Suppose a new category appears at a particular moment in time. In that circumstance, we can either reorganize the previous observations according to the new categorization, assume that none of the previous (posterior) observations belong to the newly created (eliminated) category, or introduce an interpolation.

4.1 CPC

Much has been discussed about how patent offices prepare to tackle a sudden increase in patents due to the rapid development of emerging technologies. For instance, evidence suggests that public and private investment in R&D in nanotechnology led to a significant rise in related patent applications (Absalom et al., 2006; Angelucci et al., 2012). Absalom et al. (2006) mention that the European Patent Office (EPO) developed a new taxonomy in the early stages of nanotechnology R&D to ensure technical expertise within examiners and legal certainty to innovators. In this way, patents contain information about the birth and development of technologies.

One difficulty of reliably identifying whether a TT has occurred is to ensure consistent patent classifications throughout the period of study (Lacasa et al., 2003). On October 25, 2010, the EPO and the United States Patent and Trademark Office (USPTO) announced the CPC as a joint effort aimed at establishing a harmonized classification system for patent documents; the CPC is based on the International Patent Classification system, managed by the World Intellectual Property Organization.^[4]

The CPC is divided into nine sections (A-H and Y), where each section is divided into classes, subclasses, groups, and subgroups, containing approximately 250,000 classification entries.^[5] In our illustrative case, we use patents with at least one Current CPC Class within the subgroups of the Y02E 10 group of “Energy generation through renewable energy sources”^[6] (Section 4).

One important peculiarity of the Y-section is that the allocation of new patent applications to the relevant technology code(s) is done automatically by search strategies without affecting existing classifications outside it; thus, classifiers’ intervention is constrained to developing and updating those strategies (EPO & USPTO, 2015). Furthermore, within the Y-section results, there has been a dedicated effort to reclassify all applicable patents, retroactively, with meticulous work and continuous tagging by the various patent offices, ensuring that the Y02E 10 subgroups contain only REG-related patents.

In this manner, we take advantage of the categorical property of the subgroups within the Y02E 10 group of “Energy generation through renewable energy sources” from the Current CPC Class, which contains only REG-related patents. Thus, each of the seven subgroups of the Y02E 10 group, also called categories, belongs to a different renewable energy source (e.g., wind or from the sea). We use categories that are known beforehand, where patents are sorted into seven separate categories. All patents classified under the “Wind” category are useful to generate energy using the wind but are sufficiently different from those “From the sea.”^[7]

Climate change mitigation technologies (CCMTs) involve several areas of knowledge. Thus, technologies using the existing alternative patent classifications can lead to an incomplete retrieval of documents. As stated by the EPO, “Y02 is a tagging scheme that enables documents relating to sustainable technologies to be retrieved quickly and accurately across classification categories.”^[8]

4.2 The Data Set

We illustrate the implementation of TDV using public patent data provided by the USPTO. In particular, we consider patents with at least one Current CPC Class^[9] within the Y02E 10 group of “Energy generation through renewable energy sources.” The Y02 class aims to identify CCMTs as stated in the Kyoto Agreement (Angelucci et al., 2012), which contains the Y02E subclass, “Reduction of greenhouse gas emissions, related to energy generation, transmissions or distribution.”

We searched all patents within each subgroup of Y02E 10 group (one-dot level, two-dot level, and three-dot level).^[10] Each subgroup contains materials, systems, or structures used in energy generation depending on the renewable source.^[11] For instance, in photovoltaic (PV) energy, the one-dot level is Y02E 10/50 “PV energy,” a two-dot level is Y02E 10/54 “Material technologies,” and a three-dot level is Y02E 10/545 “Microcrystalline silicon PV cells.”

Our database consists of 39,236 patents filed from 1973 to 2015. Although there is updated information on patent applications, we decided to exclude patents filed from 2016 onward from the analysis to avoid false TTs due to unobserved recent patent applications. This phenomenon exists because in Article 1122 of the Manual of Patent Examining Procedure, the USPTO allows for the possibility of requesting an application to be non-publishable until its acceptance, so we may not have been able to publicly observe the accurate quantity of recent applications. Considering patents granted up to January 2020,^[12] the mean lag between the issuance and their filing date is 2.9 years, with a standard deviation of 1.5 years. Therefore, we decided to exclude patents corresponding to the period close to the mean lag plus one standard deviation.^[13] Note that the mean lag in our database is close to 2.4 years obtained by Popp et al. (2004).

A patent can have more than one Current CPC Class, but we analyzed only those patents recorded with at least one of the REG categories (Table 1). When all subgroups within REG belong to only one subgroup, we assigned a patent to that category. However, when the subgroups belong to two categories, these patents appear in two categories; the same principle applies to patents with three or more subgroups.

In Table 1, we show the seven REG categories within the Y02E 10 group and the complete list of Current CPC classes within each one.

Figure 1 shows the number of patents by REG category depending on the year they were filed. We observe that the mix of REG sources (subgroups) has changed since the 1970s. During the 1970s, the predominant source was solar thermal, followed by PV. In the early 1980s, PV patents surpassed solar thermal ones, and these two sources remained relevant for the next two decades. At the beginning of the 2000s, the wind source category dramatically increased, exceeding solar thermal but remained below that of PV, which continues its rising trend. Since 2011, a general decline in patenting is observed. This phenomenon may be related to the maturity of some REG technologies, such as the PV and wind (International Energy Agency, 2020).

Figure 1

Number of patents filed by REG category.

4.3 TTs’ Identification

To use TVD, we need the number of periods within the whole sample and the distance between them. Thus, we define the periods using two criteria, natural periods and PSNO, which we describe in the following paragraphs.

We first divide our data by natural years, each constituting a “natural period”; then, we select the distance of comparison between natural periods. The distributions may look very similar when periods are too short and the distances too close. Therefore, shorter periods may need further distance calculations. We test a variety of combinations of years and periods of distance. Interestingly, we obtain high consistency in defining two TT moments in the REG case: one in the 1980s and another between the end of the 1990s and the beginning of the 2000s.

In Figure 2, we show different year-windows and the distance between them. If we take 1-year periods and calculate the distances between five periods, we identify two peaks (Figure 2(a)). The first peak compares the distribution in 1981 with one in 1986. The second one signals a peak in the distance between the distribution in 1997 and the one in 2002. The graph with 2-year periods with comparisons of two period distance (Figure 2(b)) exhibits more clearly the two peaks: the first one compares the distribution of 1981–1982 with 1984–1985; the second one compares the 1998–1999 distribution with that of 2002–2003. In Figure 2(c), we again observe the two peaks in the 3-year periods with consecutive (distance 1) comparisons: 1983–1985 with 1986–1989, and 2000–2002 with 2003–2005. The 5-year periods with consecutive comparisons, Figure 2(d), give a similar picture, with a first peak in 1978–1982 compared with 1983–1987, and the second comparing 1998–2002 to 2003–2007.

Figure 2

TVD for natural periods: (a) 1-year periods with five-period distance, (b) 2-year periods with two-period distance, (c) 3-year periods consecutive, and (d) 5-year periods consecutive.

For illustration purposes, in Figure 3, we show the histograms for 5-year natural periods comparing 1978–1982 and 1983–1987 (Figure 3(a)) and 1998–2002 and 2003–2007 (Figure 3(b)), which are associated with the two red points of Figure 5(d). The most notorious change from 1978–1982 to 1983–1987 is a relative reduction of subgroup 4 (solar-thermal) and an increase in subgroup 5 (PV). During the second peak, from 1998–2002 to 2003–2007, we observe a relative reduction of subgroups 4 and 5, which transition to an increase of subgroup 7 (wind).

Figure 3

Distributions during TT for 5-year natural periods: (a) distributions during 1978–1982 (red) and 1983–1987 (yellow) and common (orange) and (b) distributions during 1998–2002 (red) and 2003–2007 (yellow) and common (orange).

We repeat the exercise for PSNO Figure 4. In this way, we correct for a variability bias due to the greater number of patents being registered in recent years. Similar to the case of natural periods, a greater number of periods would result in shorter periods, which would demand a comparison conceding a greater distance.

Figure 4

TVD for PSNO: (a) 40 PSNO consecutive, (b) 40 PSNO with three-period distance, (c) 30 PSNO consecutive, and (d) 25 PSNO consecutive.

If we consider 40 PSNO and a consecutive comparison (Figure 4(a)), we find two notable peaks: one between June 1982 and March 1988,^[14] and the second between March 2000 and April 2003.^[15] For the same 40 PSNO and a comparison of distance 3 (Figure 4(b)), we obtain the same two peaks: one between June 1979 and March 1988,^[16] and the second, between January 1995 and April 2003.^[17] For 30 PSNO with a consecutive comparison (Figure 4(c)), we obtain two main peaks between October 1980 and December 1986,^[18] and between June 1999 and October 2003.^[19] For 25 PSNO (Figure 4(d)), the identification is less precise: one between October 1980 and December 1991^[20] and the second between October 1997 and August 2004.^[21]

For robustness, we performed a multinomial chi-square test of independence between each pair of distributions to test whether these two come from the same distribution.^[22] If we assume that the realization of patents in a particular year is a “sample,” in the absence of technological change, the relative distribution of patents among the categories must be the same between the two periods.

Thus, by performing the independence test between two samples, we are testing whether these two belong to the same population, and when we reject this null hypothesis, there has been technological change. However, as we explained earlier, we are interested in finding those moments with a drastic technological change; thus, if the test statistic is much larger than the critical value, then it would signal a TT.

Since the chi-square test of independence is affected by the size of each sample, we only performed it on the distributions grouped in natural years. Also, we did not have extreme values for the frequency distributions of most of the natural (and synthetic) periods analyzed; thus, this multinomial test performed satisfactorily, and we found no need to introduce additional tests (see Cai & Krishnamoorthy, 2006).

In Figure 5, we offer a comparable graph for Figure 2 of the chi-square statistics for every paired of distributions. Also, we show two levels of significance α = 0.01 (solid blue line) and an unusual α = 0.001 (dotted blue line), due to the fact that even at those demanding significance levels, we rejected the hypothesis that the compared distributions belong to populations with the same frequency distribution. That is, for most comparisons, there is technological change but not necessarily a TT, which is consistent with our findings with the TVD approach.

Figure 5

Multinominal chi test: (a) 1-year periods with five-period distance, (b) 2-year periods with two-period distance, (c) 3-year periods consecutive, and (d) 5-year periods consecutive.

The TTs identified correspond very consistently to local maxima of the statistic. In the first graph, the first TT, at the pair 1981–1986, coincides with a local maximum of the statistic with a value of 77.2952 that corresponds to a p -value of 1.29 × 1 0 − 14 ; the second, at the pair “1997–2002,” gives a local maximum of the statistic with a value of 55.4242 that corresponds to a p -value of 3.8055 × 1 0 − 10 . The correspondence is maintained for the different considerations of periods and distances. It must be noticed that since the statistic is sensitive to the size of the sample, it shows a third local maximum for the pair 2008–2013. However, we did not find a clear signal of a TT with the analysis of the TVD under different configurations.^[23]

In this section, we show that the TVD approach has consistently identified two TTs, despite considering a wide range of period sizes and comparison distances. Note that since we have not considered any external information, the measure is ideal for an initial exploratory analysis and only uses the relative frequency of each category within REG field.

4.4 Interpretation

Our methodology systematically indicates two moments of notable change in the distribution of patents for REG. Although TTs could have occurred naturally, we aim to acknowledge some policies linked to the two moments identified through the TVD calculation, which qualitatively reinforce the correct identification of the TTs. The first peak in the early 1980s could result from the research that would have started in the late 1970s. Between the late 1990s and the beginning of the 2000s, the second peak could relate to the Kyoto Treaty. However, we do not pretend to establish any causality between the policies mentioned and the TTs.

The oil embargo of 1973 triggered interest in alternative energies, which motivated a wide range of energy conversion techniques from renewable sources. Among these, PV panels started to be considered a viable option for commercial use (Sørensen, 1991). The oil crisis in the 1970s triggered public funding for R&D programs aimed to advance PV generation, mainly from Japan and the USA, and it is calculated that around 60% of the cost reduction in this technology was due to public and private R&D (International Energy Agency, 2020). President Nixon’s Project Independence in 1973, President Ford’s Energy Policy and Conservation Act in 1975, and the Public Utility Regulatory Policies Act in 1978 also promoted renewables and have been related to technological changes that led to cost reductions in wind and solar PV (Clayton, 2004; Smith, 2004; Solomon & Krishna, 2011).

In 1997, the Kyoto Protocol was adopted by almost 200 countries, committing the signers to reduce their carbon emissions by 2012. Evidence suggests that the protocol triggered a rapid increase in patenting activity in CCMTs. A positive effect has been found on applications for renewable technology patents in countries with emission targets (Miyamoto & Takeuchi, 2019). This increase has significantly affected solar and wind technology patents in European countries (Johnstone et al., 2010). Even though the USA signed the protocol on November 12, 1998, the treaty was never ratified by the US Senate, as needed.

However, the protocol could have awakened economic incentives in other countries that had signed and ratified the protocol, which led to increased interest in patenting in the US. Also, US polls on political concerns and political initiatives, such as President Clinton’s tax on BTUs proposal, or the Climate Change Action Plan, indicated potential changes in energy regulation in favor of renewable sources. Chalvatzis et al. (2020) argue that the Kyoto Protocol redirected innovation efforts towards REG, promoting the interest of the corresponding R&D to meet the growing demand in that sector. The prevailing dominant positions in the patenting of PV and wind could reflect the extensive use of these technologies in the marketplace (UNEP et al., 2010).