The main purpose of this article is to analyze the current EV market in Taiwan. The article will be split into 3 parts :
The following results are obtained by statistical analysis and model. This article will be provided as reference and evidence for PPO department to conduct long term research, decision making, analysis, etc.
- Note : Case A uses time series data from 2011 to 2021 and focus on trends and values. On the other hand, case B uses only 2021 data and focus on current EV environment development of each country.
In case A, the aim is to find out countries that have similar trends
and percentages of EV sales share and
EV stock share from 2011 to 2021. Countries that have
similar curves to Taiwan in the past 10 years can be observed, and these
countries may be considered as references for future analysis for
Taiwan.
Most of the data for case A study were obtained from the IEA ( International Energy Agency ) website. The following table are the explanations of the data variables.
For example, the data of Australia :
After rearranging the data by EV sales share and
EV stock share, we then obtained the following table for
further clustering analysis :
Time series cluster analysis for EV sales share and
EV stock share will be conducted respectively by using R
package dtwclust and its tsclust() function
with “DTW Distance”. The data was normalized while pre-processing, and
both trend and value of each country were majorly considered while
clustering.
We have observed the trends and values of EV sales share and EV stock share in each country. Thus more detailed market investigations target on these countries can be carry out. We can also use the result for further market predictions or new car model introduction decisions. EV car markets and consumer behaviors in Australia, Southern Europe, and few countries in Latin America can be consulted for further studies.
In case B, the aim is to find out that among the 27 sample countries, which countries have similar EV environment development. 10 variables were selected to present the current EV environment development of a country. The 10 variables were classified into 3 categories : EV charger constructions, EV related prices, and automotive industry development levels. By analyzing the variables, the similarity of EV development in each country can be observed.
Most of the data for case B study was obtained from the IEA ( International Energy Agency ) website. In order to present the latest EV environment development of each country, only 2021 data were used. The following table are the explanation of the variables in the dataset.
Below is the data for cluster analysis :
Hierarchical Clustering with “Euclidean Distance” and “Wards Method”
will be conducted by using R base function hclust(). The
data has been normalized while pre-processing.
“EV charger constructions”, “energy prices”, “EV car prices”, “automotive industry development levels”
The clustering splits all 27 countries into 10 groups, Taiwan is labeled in the dark-blue rectangle area. In the category of EV charger constructions which considered fast and slow chargers, Taiwan was grouped with Japan and Portugal, following up with countries mainly in Southern Europe and Southern America. From the perspective of current EV charger availability and density, Taiwan is more similar to Japan, Southern Europe and Southern America countries.
The average number of EV cars that a single fast charger need to serve in each country. For instance, the value in Taiwan is 33.62681, this implies that a single fast charger need to serve roughly 34 EV cars that is in use. Small value implies that there are relatively adequate fast chargers for current EV car usage in the country. Chile, Korea, China, Poland, and Japan are the top 5 on the list, and Taiwan is in the 9th place among the 27 countries. Most EV advanced countries have larger fast charger share values since the amount of unit in use EV cars are significantly larger.
The average number of EV cars that a single slow charger need to serve in each country. For instance, the value in Taiwan is 5.98774, this implies that a single slow charger need to serve roughly 6 EV cars that is in use. Small value implies that there are relatively adequate slow chargers for current EV car usage in the country. Korea, Chile, Greece, Netherlands, and Finland are the top 5 on the list, and Taiwan is in the 8th place among the 27 countries. We can observe that availability of fast and slow chargers in Korea and Chile are significantly better comparing to other countries on the list.
The average number of fast chargers per square kilometer in each country. For instance, the value in Taiwan is 0.01559, this implies that there are average 0.01559 fast chargers per kilometers in Taiwan. Large value implies that fast chargers are relatively dense in the country. Korea, Netherlands, China, Switzerland, and UK are the top 5 on the list, and Taiwan is in the 11th place among the 27 countries.
The average number of slow chargers per square kilometer in each country. For instance, the value in Taiwan is 0.08755, this implies that there are average 0.08755 slow chargers per kilometers in Taiwan. Large value implies that slow chargers are relatively dense in the country. Netherlands, Korea, Belgium, Switzerland, and UK are the top 5 on the list, and Taiwan is in the 8th place among the 27 countries. We can see that government mostly invest in the construction of fast chargers. Also, both density of fast and slow chargers in Korea are significantly high comparing to other countries on the list.
The clustering splits all 27 countries into 10 groups,Taiwan is labeled in the dark-blue rectangle area. In the category of automotive industry development level which considered import and export of cars, Taiwan was grouped with Denmark, New Zealand, Chile, Iceland, Greece, Portugal, Finland, India, and Sweden. Most of these countries have small car market size and aren’t famous for their auto industry.
The amount of value by exporting cars ( USD ) in 2021. Germany,
Japan, USA, Korea, and Mexico are the top 5 on the list, these countries
are famous for their automotive industry. On the contrary, it isn’t
suprisng that Taiwan is in the 20th place among the 27 countries. This
export_2021 variable is included in order to consider
whether a country is more capable to increase the available choices of
new EV cars on the market.
The amount of value for importing cars ( USD ) in 2021. USA, Germany,
China, France, and UK are the top 5 on the list. Taiwan is in the 19th
place among the 27 countries. This import_2021 variable is
included because the size of the car market in each country may also be
a crucial factor for EV environment development.
The clustering splits all 27 countries into 10 groups, presents by 10 different colors.Taiwan is labeled in the dark-blue rectangle area. The result combines information from all 3 categories above. Thus from the 2021 data, we can observed that Italy, Spain, France, Belgium, and UK currently have similar EV environment development.
We can use this result for further analysis. Detailed investigation of these countries’ EV constructions and car market can be carry out. Since the result of the final clustering for all categories may be non-intuitive and hard to interpret, it is easier to focus one single categories and conduct further study. Japan, Korea, and Northern Europe countries can be investigated for EV charger constructions, EV related Prices, and automotive industry development respectively.
In case C, the aim is to forecast the 2030 passenger EV cumulative sales in Taiwan. The famous Bass Diffusion Model is used for our prediction.
Most of the data for case C study was obtained from PPO department’s
sales table. The data includes sales number ( units ) of EV, HEV, ICE
vehicles in each quarter from 2004 Q1 to 2022 Q2. For example,
time variable 2022-4 implies year 2022 quarter 4.
EV, HEV, and ICE imply are the
sales number in that quarter.
For further examination, the following plot is the visualization of
the data above ( the time variable is simplified to year
only ).
The current car market is still dominated by ICE vehicles. Due to the financial crisis in 2008, sales number had fallen to the bottom around 2007 to 2009. After the crisis, sales number started to rise and has remained steady around 400000 units sold per year since 2014. However, the number never reaches 2005’s 500000 units, which is the peak over the past 20 years. We can also see that there is a tendency of decreasing in ICE vehicles sales after 2020, mainly because of COVID-19 and chips shortage. This may become the divergence point for ICE vehicles and alternative fuel vehicles. Sales numbers of EV and HEV start to rise along with more and more consumers begin to accept new technologies. It is just a matter of time for alternative fuel vehicles to replace ICE vehicles.
Since the plot above is dominated by the sales number of ICE vehicles, the trend of EV vehicles is hard to observe. Thus plot exclusively for EV sales number is shown below.
Sales share in Taiwan has increased dramatically from 0.21% to 2% since 2018. Before 2018, the percentages of sales share were practically negligible and increased very slowly. After 2018, sales share of EV cars suddenly boomed, this resulted from the introduction of Tesla Model 3. Furthermore, a significant amount of passenger EV cars, mostly CBU, has been introduced to the Taiwan car market since 2019. For instance : Porsche Taycan, Jaguar I-Pace, Hyundai Kona/Ioniq EV, Audi e-tron, etc. Along with the increase of available choices of new EV cars in the market, it is expected that the EV sales share in Taiwan will continue to rise in the near future.
The Bass diffusion model was developed by Frank Bass. It consists of a simple differential equation that describes the process of how new products get adopted in a population. The model presents a rationale of how current adopters and potential adopters of a new product interact. The basic premise of the model is that adopters can be classified as innovators or as imitators and the speed and timing of adoption depends on their degree of innovation and the degree of imitation among adopters. The Bass model has been widely used in forecasting, especially new products’ sales forecasting and technology forecasting.
Original Bass model :
\[f(t) = m\frac{(p+q)^2}{p}\frac{e^{-(p+q)t}}{(1+\frac{q}{p}e^{-(p+q)t})^2}\]
However, the original Bass model does not take the restriction of supply and demand, the aging of technology, and the natural replacement of vehicles into account. Therefore, it was decided to use the solution of the form which represents the current total number of EV on the market :
\[F(t) = m\frac{1-e^{-(p+q)t}}{1+\frac{q}{p}e^{-(p+q)t}}\]
Parameters :
< Left : Cumulative Sales / Right : Current Sales ( Figures from Wikipedia ) >
The parameters \(P\) and \(q\) were obtain by non-linear square fitting. We assumed that EV will be accepted by every consumers on the market in the end. Thus the market potential parameter \(m\) was assumed to be the total market size which is roughly 4 million. ( Bass model does not take into account the natural replacement of vehicles. The cycle of consumers replacing their cars is about 10 years. Sales data from the past indicates that there were about 400,000 units sold annually. By combining the previous information, we assumed the potential market size to be 400,000 * 10 = 4,000,000)
- Note that there are several assumptions needed in the forecasting process, and result of the prediction might change along with different values of parameter \(m\), thus the value of \(m\) needs more indepth study if necessary.
The model is trained by using cumulative EV sales data from 2010/Q3 to 2020/Q4, and data from 2021/Q1 to 2022/Q4 is used for testing the model performance. Following tables are the training and testing data set.
We have tested several models with different parameters. After checking normality of the residuals and the model parameters’ significance, we then compared each model’s testing performance ( prediction of data from 2021/Q4 ~ 2022/Q2 ) by using AIC criterion. Smaller value of AIC implies better model performance. AIC of the final model is 658.0509, which is the smallest among all the models we have tested. The final model has parameters \(p = 0.0000007996\), \(q = 0.15\), given the assumption of \(m = 4000000\). This implies that EV products have a really low innovation effect. However, the imitation effect seems relatively normal since typical range of technology products have range between 0.3 and 0.5. We can conclude that after the starting phase of EV, the adoption rate will relatively increase faster.
The fitted curve and its testing result are shown in the following plot.
Data from 2010/Q3 ~ 2020/Q4 are the training set for parameters fitting. Data from 2021/Q1 ~ 2022/Q2 are the testing set for evaluating model prediction ability. The black dots are the cumulative sales. The black curve is the fitted model. The blue areas are the 90% confidence intervals of the predictions.
Performance of the model in the testing data set is pretty good since most of the data points are included in the 90% confidence intervals. Furthermore, the slope of the increasing trend in the last 3 quarters are predicted correctly. On the contrary, model performance in the training data set seems more inaccurate. However, this implies the final model is more generalized comparing to other models which tend to overfit the training data. We can see that the slope of cumulative sales becomes more gentle after 2020/Q1 and 2021/Q2, and our final model has taken this information into consideration. Thus the prediction of the trend and slope after the sudden increase of cumulative sales are reasonable.
Since the market potential \(m\) is assumed to be 4 million and supposing all potential consumers will eventually buy EV, the above result can be interpret as : Until 2029 Q4, there will be \(36\% \pm 14\%\) electric vehicles being sold among the total new passenger car market size in Taiwan (4,000,000) . However, this result do not take charging restriction problems into account.
The following is the time prediction table when the cumulative sales reaching 5%, 10%, 15%, and 20%.
According to the government statistics report, about 68% of the people have their own private parking space. Rest of the people rent parking space or just park their car on the roadside. However, whether having a private parking space has great impact on setting up an EV charger. In fact, EV charger installation is the primary consideration when buying an EV for most of the consumers. The main barrier that obstruct the widespread of EV in Taiwan is ascribe to the lack of private parking spaces and the resistance of apartment committee. Therefore, the above prediction won’t make any sense if we can’t overcome this barrier.
As mentioned previously, the EV total market potential is assumed to be 4 million. However, the final total market potential can’t be reached due to parking space and charging problems. Thus the prediction result may be truncated. For instance, in Figure 1, the upper limit of potential consumers is truncated to 1904000 instead of 4 million. The number 1904000 is obtained from having 68% of the total 4 million potential buyers own private parking space, and supposing that 70% within them are able to set up EV chargers in their parking space ( 4000000 * 0.68 * 0.7 = 1904000 ). By replacing the assumption of 70% in figure 1 to 50% and 20% respectively, we have the results shown in Figure 2 and 3. Therefore, we can observe different truncated prediction. Figure 2 implies the prediction will be truncated around 2030, and figure 3 implies the prediction will be truncated around 2026. After the period of truncation, the curve may increase and reach 4 million after charging problems are solved or eventually decrease to 0 since EV die out due to the inconvenience. Development of the curve are highly correlated to several variables including government policies, technology improvements, consumer demands, etc.
Below is the Rogers diffusion innovation plot. With successive groups of consumers adopting the new technology (shown in blue), its market share (yellow) will eventually reach the saturation level. The blue curve is broken into sections of adopters. The Rogers model usually assumes that, if a market share of approximately 12% ~ 16% is achieved, it means that the innovation has a very good chance of being adopted by the entire market. However, EV is still in the innovators stage in Taiwan. This may contradict our previous assumption that EV will be accepted by every consumers on the market in the end.
< Rogers’ diffusion innovation >
One of the researchers exploring Rogers’ concept was Moore. Moore, in his famous publication Crossing the Chasm, discussed the gaps and barriers that must be overcome to move from one customer group to the next. The largest gap, which he calls “the chasm” ( shown in the plot below ), exists between the early adopters and the early majority. Since the current EV market in Taiwan is still in the innovators stage and hasn’t meet the chasm, predictions and forecasting results will be influenced at any time in the future.
< Moore’s Crossing the Chasm >
The increasing level of EV sales will become more significant throughout the time. Introduction of new EV models is necessary. There is a large potential market for electric vehicles waiting for companies to invest. From the increase of EV adoption rate, it shows that EV isn’t an unreachable high technology product to consumers anymore. Not only innovators but also imitators start to consider buying EV as their next car due to the rising visibility on the road.
However, charging problems must be solved in the first place. EV charging system in the urban area may require lots of effort and time to become more complete, thus the promotion and advertisement of EV can first target on the consumers that live in the countryside or suburb. Most of these people have private space for parking and can easily set up their own EV charger comparing to those who live in the urban area.
All in all, since the development of EV charging infrastructure and system currently isn’t promising, introducing large quantities of EV isn’t the primary goal at this stage. The main goal is to understand the demands of the potential consumers in the future EV era.