2. Literature Review
There are many kinds of method to forecast vehicle ownership, and we will summarize these methods in the literature, and then talk about more about the approach of Gompertz function, including the expansion of economic factors and the method improvement. Since we use GDP per capita as the dependent variable in Gompertz function, we will review the methods to forecast the growth rate of GDP.
The commonly used prediction method for vehicle ownership is to correlate vehicle ownership with population or economic indicators. Time-series models and econometric models related to income and travel characteristics can be used in developed countries because of the relatively stable growth of vehicle ownership and access to complete statistical data. However, only a small part of available data, such as vehicle ownership, population and GDP, can be used for modeling in developing countries [19
]. There are several methods to predict vehicle ownership all over the world. First, some researchers use Gompertz, logistics and bass diffusion models with a time trend as the independent variable [5
], and these simple and clear models describe how the vehicle ownership changes with time, however, the elastic coefficients of these models are fixed, which is not realistic. The second kind of method is the neural network method, which has high prediction accuracy and can be used for multi-factor and multi-objective analysis, but it is almost a black-box prediction, and it is difficult to explain the growth mechanism [17
]. The third method is based on the econometric model, which can predict vehicle ownership through multiple indicators related to economic development, but is only suitable for short-term predictions, and requires a large sample [21
]. The final method is based on the Gompertz model, which is widely used because it can fit the relationship between economic variables and vehicle ownership better while also having saturating level.
The economic factor is the key variable in the Gompertz model. Most researchers use the income of residents [24
] or GDP per capita as economic variables [8
] in the Gompertz model. On the choice of economic variables, because there is a deviation in the process of data statistics of income, and in some areas, the government will set some policies for vehicle market for reasons of social welfare maximization, such as purchasing limitations and other restrictions, which means that the residents cannot buy a car even their income reaches a certain level. So, economic indicators that adjusted by government regulation are more suitable for explanatory variables [8
]. Some researchers use compounded or adjusted economic factors to forecast vehicle ownership. The author of [15
] use the Gompertz model with the independent variable as a factor extracted from GDP per capita and disposable income per capita to forecast the vehicle ownership in Beijing and Shanghai. In addition, vehicle price and disposable income per capita are used. The author of [25
] simulates private car ownership on an income-level basis, takes into account car purchase prices, separates sales into purchases for fleet growth and for replacements of scrapped vehicles, and examines various possible vehicle scrappage patterns for China. Some researchers also think about the diffusion process of new technology, as well as intrinsic and extrinsic motivations, and the crowding-out effect on consumers’ purchasing decisions, in order to analyze how economic factors affect vehicle consumption [30
Besides the selection improvement of the economic factor in Gompertz model, the methodology of the Gompertz model is improved in some research as well. Lu et al. [31
] improved the stochastic differential equation related to the Gompertz curve, so that the model can present the remaining slow increase when the S-shaped curve has reached its saturation level and can better fit the real data when there are fluctuations in it. Lian et al. [32
] utilizes data-driven symbolic regression to automatically find a generalized function by symbolic regression (NE-SR) for passenger car ownership, and then uses the new proposed function to forecast passenger car ownership in China. Dargay et al. [26
] builds a model that explicitly models the vehicle saturation level as a function of a country’s urbanization and population density characteristics based on the Gompertz model.
The most commonly used economic factor is GDP per capita, and many researchers focus on the better prediction of GDP per capita when they use the Gompertz model. There are several methods to predict GDP per capita. The first one is the analogy method. Some researchers calculate the average annual growth rate of GDP using past data, and then use this growth rate to predict future GDP [15
]. Time series models of past GDP growth rate, such as a multiple autoregressive moving average (MARMA) model, are also used to predict future GDP [18
]. Some researchers also reference the experience of other countries to predict China’s future economic growth [34
] because they believe that countries in a similar development stage share a similar growth rate. This analogy ignores differences in population structure and in institutions [16
]. The second method is the growth accounting method, which predicts future economic growth based on the potential growth of total factor productivity, capital and labor respectively [35
]. However, this method relies on the form of production function and the output elasticity of capital and labor. Other researchers use modern macroeconomic models, such as the CGE (computable general equilibrium) model or the DSGE (dynamic stochastic general equilibrium) model to predict GDP [8
], which simultaneously consider the linkage of multiple variables of the demand side and the supply side. However, it is too complex and difficult to compare the results of different models, because the mechanisms between variables are not easy to see. Some researchers considered the impact of population structure on economic growth [36
]. Based on this idea, Bai and Zhang [16
] believed that although the investment and human capital structure of China is similar to that of Japan, Singapore, South Korea, and Taiwan were in the same development stage. The employment rate in the total population of China is different from these countries, so they used the analogy method to predict the growth rate of China’s GDP per labor based on the growth model of GDP per labor in these countries, and added the growth rate of China’s labor force to obtain the growth rate of total GDP. This method not only captures the common convergence and avoids the specific form of production function, but also considers the uniqueness of the growth rate of China’s labor force. However, for the sake of simplification, this analytical framework assumes that different economies have similar growth patterns at similar stages of development, which ignores the important role of other factors, such as savings (investment) rate, human capital, technological innovation and institutional factors, in economic growth and convergence, which may have a certain impact on the rationality of the forecast results. Some policies that can intense the decrement of employment include, for instance universal two-child policy and delay the retirement policy are not included in the estimation process. Anyway, this method is an attempt to consider demographics.
5. Factors That Affect the Trend of Vehicle Ownership in Each Province
The ranking of vehicle ownership of most provinces in the future will change and the speed and extent of growth varies across provinces. There are several features in the growth process: (1) Higher-ranking provinces will tend to decline in the future, while the lower ones will tend to increase in the future; (2) the current first-tier provinces, such as Beijing, Shanghai, Tianjin, and Guangdong will fall into the low-level vehicle ownership group in 2050. We will analyze several causes that lead to these results.
5.1. Differences in the Growth Model of Vehicle Ownership
The vehicle growth model has a great impact on the growth speed and the ranking of vehicle ownership in each province. Just as Table 2
shows, the model with rapid growth rate can partially explain why some provinces can keep their high rankings in the next 30 years, such as Inner Mongolia, Zhejiang, Jiangsu and Shandong, which is shown in Table 2
The average-speed growth model leads to the unchanged ranking of some provinces with middle-level vehicle ownership now, such as Jilin, Liaoning, Shaanxi, Hainan, Gansu and Guangxi. The slow growth rate of the vehicle ownership model can partly explain why some provinces have declining rankings. For example, the development model with low-growth rate, which is only 0.289, is applicable to Beijing, Shanghai and Tianjin. The growth-rate parameter of Guangdong’s model is 0.671, which is still below the national average. Similarly, the slower growth rate in Tibet, Yunnan, and Shanxi provinces also led to the decline in their rankings. The faster vehicle growth model is more likely to lead to increasing ranking, such as Fujian Province, who will rise to the high vehicle ownership group from the middle group now.
5.2. Differences in GDP per Capita and Its Growth Rate
Another important factor in the Gompertz model is GDP per capita. We also group each province into high (labeled by 1), medium (labeled by 2) and low (labeled by 3) groups according to their ranking of GDP per capita. Figure 7
shows the group number of each province. It can be seen that the ranking of GDP per capita in each province is basically consistent with the ranking of vehicle ownership, especially before 2025. This result shows that GDP per capita also has a great impact on the vehicle ownership of each province. The ranking of GDP per capita of each province changes after 2025. Some provinces in the middle group now will jump to the high-level group, such as Chongqing and Henan, or fall into the low-level group, such as Xizang, Shanxi, Guangdong and Liaoning. The ranking of some provinces in the low-level group now will increase, such as Sichuan and Hunan, which explains the rise in their vehicle ownership rankings. Figure 8
shows the detailed changes of ranking of GDP per capita of each province.
shows the growth rate of GDP per capital of each province in different years. The province order is sorted by the 2018 value. It can be seen that there exist stepwise changes from the lower left corner to the upper right corner, which is consistent with the distribution trend of GDP per capita, and current GDP per capita is negatively correlated with its growth rate. For example, the growth rate of GDP per capita is only about 5% in rich province like Beijing and Shanghai, and it will gradually decline to around 3% in the next 30 years. The value in Tianjin and Jiangsu will decline to around 3% from around 6%, and Henan, Hunan, Guangxi and Guizhou will have the highest values, which is about 5% in 2050. The growth rate of GDP per capita will decline from the range of 7–13% to the range of 3–5% in 2050 in other provinces. The difference in the growth rates of GDP per capita across provinces ultimately led to differences in GDP per capita, which in turn will affect vehicle ownership.
Vehicle ownership is more likely to increase rapidly in provinces with a higher growth rate of GDP per capita under the same development model. For example, in Henan, Anhui, Sichuan, Jiangxi and Hunan provinces, the GDP growth rate per capita is more than 10% in 2018, so it will converge slowly in the next 30 years, which can lead to rapid growth in GDP per capita, and thus rapid growth in vehicle ownership. This phenomenon also exists in the three provinces of Chongqing, Hubei and Heilongjiang, whose growth rate of GDP per capita is 9%. The growth rate of GDP per capita is around 8% in the provinces like Shaanxi, Jilin and Shandong, whose vehicle growth rate is also on the average level, so that their ranking of vehicle ownership will not change much in the next 30 years. The ranking is likely to decline in provinces with a middle or low-level growth rate of GDP per capita and low growth rate of vehicle growth model, such as Guangdong, Xinjiang, Ningxia, Qinghai, Shanxi, Hebei, Hubei, especially Beijing, Shanghai and Tianjin. While the ranking may increase in provinces with low GDP per capita growth rates but rapid growth model of vehicle, such as Jiangsu, Inner Mongolia and Zhejiang. The growth rate of GDP per capita is 7% in these provinces, but the suitable vehicle growth pattern is the Italy and Spain model, which has a high growth rate, so their vehicle ownership is always high ranked.
5.3. The Growth Rate of GDP per Labor and Employment Proportion
The growth rate of GDP per capita can be decomposed into the growth rate of GDP per labor and the growth rate of the employment rate. Therefore, we compare the two factors across provinces.
The estimated result of parameters in Equation (5) is μ
= 2.427%, θ
= 0.673. We use the year 2014 as the starting point to obtain the growth rate of GDP per labor of each province by iteration from 2018 to 2050. The results are shown in Table A2
. The order of the provinces is sorted by the 2018 value. It can be seen that from the lower left corner to the upper right corner, the growth rate of GDP per labor is similar to the growth rate of GDP per capita, which also presents step changes, indicating that the growth rate of GDP per labor is a major factor determining the growth rate of GDP per capita. The growth rate of GDP per capita presents more tidy and rigorous step changes, which is consistent with the economic convergence theory. However, due to the different development stages, the average GDP growth rate is different in each province, and the growth rate of GDP per labor is more dispersed than the growth rate of GDP per capita in 2018.
A high growth rate of GDP per capita leads to rapid increases in vehicle ownership. The provinces with average growth rate of GDP per labor probably keep their ranking unchanged or even drop when their growth rate in the Gompertz function is low, such as in Guangdong, Xinjiang, Ningxia, Qinghai, Shanxi, Hebei and Hubei. Similarly, the low growth rate of GDP per labor usually leads to the low level of vehicle ownership, such as in Beijing, Shanghai and Tianjin. The exception appears when the growth rate in Gompertz function is high, for instance, Jiangsu and Inner Mongolia and Zhejiang provinces, whose suitable models are Italy and Spain, and will remain in the top in the ranking of vehicle ownership.
shows the growth rate of the employment rate in each province, which reflects the difference between the growth rate of GDP per capita and GDP per labor. We can see two phenomena in this table. (1) The value of the growth rate is almost negative, and the negative value becomes even smaller in each province, which means the employment proportion decreases in the future. This is consistent with national level results predicted by [16
]. There are two reasons for this: First, the demographic dividends, which contributed a lot to economic growth in the past decades in China, are disappearing because of the aging population. The population and labor structure still maters but they begin to have a negative effect on the economic growth rate. Second, with the development of artificial intelligence and automation, many jobs are replaced by machines, so that a reduction in employment opportunities will lead to a decline in the employment rate. (2) The provinces with a lower GDP growth rate per labor, such as Shanghai, Jiangsu, Beijing and Tianjin, have lower growth rates of the employment rate, and this leads to a lower growth rate of GDP per capita and the decline of vehicle ownership rankings. The growth rate of the employment rate is higher, even positive in the provinces in the lower development stages, such as Tibet and Heilongjiang. This is because, although there are more employment opportunities in the more developed provinces, the artificial intelligence and automation penetration rate is high, so more employment is replaced by machines. More labor must migrate back to the less-developed provinces, which leads to the high growth rate of employment in these areas. Some western provinces, such as Gansu and Qinghai, cannot attract people due to their poor industrial structure. Thus, the employment in these regions will drop a lot as well in the future.
6. Regional Analysis
In order to analyze the connections and differences between provinces, we focus on the characteristics of the trend of vehicle ownership in the five regions. The Yangtze River Delta includes Shanghai, Jiangsu, Zhejiang and Anhui; the Pearl River Delta includes Guangdong, Fujian, and Hainan; the Bohai Bay region includes Beijing, Tianjin, Hebei, and Shandong; the Guanzhong-Bashu includes Shaanxi, Chongqing, and Sichuan; and the western region includes Guangxi, Guizhou, Yunnan, Tibet, Gansu, Qinghai, Ningxia and Xinjiang. Although the five regions don’t include all of China’s provinces, they are representative enough to allow us to analyze the characteristics of regions that are united by several provinces. And there are several development policies placed on these regions which can help us understand the connections in these regions. Table 5
shows the vehicle ownership forecast of each region. It can be seen that vehicle ownership trends are similar in each region. Vehicle ownership is almost around 700 by the end of 2050, which means the development among each region is similar. This character also exists in the growth rate of GDP per labor and employment proportion, which is shown in Table 6
and Table 7
. Except for the western region and the Guanzhong-Bashu region, whose growth rate of GDP per labor is a little higher, regions all start to converge from around 7% to 4% by 2050. Table 7
shows the growth rate of employment proportion in the five major regions, and we can see the largest drop of employment appears in Guanzhong-Bashu district, and the decreasing of employment degree is similar in other regions.
There are several reasons for the phenomenon. First, there are interactions between provinces in each region. Although there are differences in each province within each region, they may develop with complementarity and mutual cooperation, so there will be no large weaknesses smoothing out the regional growth rate. Second, the government will always coordinate the balanced development of large areas in its long-term development strategies. A lot of development strategies to promote balanced regional development have been announced. In recent years, the government has successively implemented coordinated development strategies in of Beijing-Tianjin-Hebei, the Yangtze River Delta, the Guanzhong Tianshui Economic Zone and the Chengdu-Chongqing Economic Zone. These development strategies are mainly achieved through industrial support policies and supporting funds, which directly affect the employment of the labor force and internal and external labor mobility in the supported regions, thus affecting the proportion of employment in the region. This will also affect the industrial layout in the region and the infiltration and application of technology in the industry, which will indirectly affect the labor productivity in the region, thus affecting the growth rate of GDP per labor in the region.
Therefore, China’s coordinating regional development strategy will not only promote the balance development of regions and achieve effective allocation of production factors within the region, but also can achieve effective allocation of elements between regions by the movement of labor and technology diffusion, and then affect the growth rate of GDP per labor and the change of employment proportion. The two effects together will affect the regional growth rate of GDP per capita, so that the vehicle development of each region will finally show a convergence trend.
The growth of vehicle production is not only the result of economic development, but also an important driving force for economic development and for industrial development. However, the increase of vehicle ownership has also triggered environmental problems, the severity of which varies in different provinces. Only by accurately predicting vehicle ownership, building a high-quality supply of vehicles, and following up sales with support services and industrial development planning can we make full use of the advantages of the vehicle industry. It is of great significance to build supporting infrastructure and other services based on accurate forecasts of vehicle ownership in various provinces because of the variance of economic development, the carrying capacity of resources, and different degrees of transport planning in each province.
We used the Gompertz model to predict China’s provincial vehicle ownership from 2018 to 2050. Considering the impact of the population structure, we summed up the growth rate of GDP per labor, the population and the employment rate to get the growth rate of GDP and then the GDP per capita of each province. We found that the vehicle ownership rate in each province will grow rapidly in the next 30 years; however, the change in the ranking of vehicle ownership among provinces varies. Then, we analyzed the causes that affect the change of ranking. On the one hand, the suitable growth pattern, which is reflected by the coefficient in the Gompertz model, has an important effect on vehicle ownership. On the other hand, the GDP per capita is another key factor that affects vehicle ownership. The stage of economic development and government policy are related to the growth rate of GDP per labor and employment rate, which then affect GDP per capita.
Therefore, we propose the following suggestions: first, the rapid growth of vehicle ownership, especially the rapid growth of traditional internal combustion engine vehicles, brings the increasing consumption of oil. Therefore, the government needs to accelerate the growth rate of new energy vehicles to replace the consumption of oil of traditional internal combustion engine vehicles. Second, from our result, the vehicle ownership will be balanced all over the country because the increasing need of undeveloped province in the future, so it is necessary for the government to make capacity planning and give some policy for the vehicle industry to ensure and induce the balanced expansion in the western area to meet the increasing demand of vehicle. Third, although the pollution problems are paid high attention by the government and society in eastern and developed provinces, the government and vehicle companies should do some preparation in some western province, because the vehicle ownership will grow fast in 30 years according to our forecast, and limit pollution within a reasonable level to better promote economic and green development.
There are some limitations in our research. First, it is impossible to divide the car ownership of traditional internal combustion engine vehicles and electric vehicles because of data availability. For example, the state has different purchase restrictions on these two types of cars in Beijing. Second, the benchmark countries we use are only OECD countries because of data availability as well. Our results will be more precise if we have more sample data of the benchmark countries or state level vehicle ownership data of U.S. or other region level data of comparable countries to China. Finally, some policies that can tense the decreasing of employment, for instants, universal two-child policy and delay the retirement policy are not included in the GDP per capital estimation process.
In the follow-up research, we will try to investigate several topics as below: firstly, since the degree of development of shared bikes or cars varies in different provinces, we will consider the influence of the penetration rate of shared travel market and the substitution and combined travel mode of shared car and private car travel on the prediction of car ownership in different provinces. Secondly, we will consider more policies that can affect the results in the estimation process, e.g. tensing the purchase restrictions, universal two-child policy and delay the retirement policy. Thirdly, we will use state level of U.S. or other region level data of comparable countries to China to estimate Gompertz function. Finally, we will do some research about the impact of vehicle ownership on other economic variables, for example, transportation, environment pollution, employment and so on, in order to prepare for the growth of vehicle ownership.