Abstract
As an important part of China’s strategic emerging industries, the development of AI industry is rapid. The state and local governments have issued policies to encourage its development. The number of AI enterprises is increasing, its quality is also improving, and the application scope and market share of products are gradually expanding. With the large amount of capital input, the efficiency of capital input-output in AI enterprises has been paid more and more attention. In this paper, 48 listed companies in AI stock market are selected, with undistributed profit, net profit and surplus reserve as output indicators, monetary capital, accounts receivable, inventory and fixed assets as input indicators, and DEA model is used to evaluate their capital input-output efficiency. The results show that the capital input-output efficiency of 16 companies is DEA effective, and for non DEA effective sample companies, relevant improvement strategies are proposed.
Introduction
AI (AI) was first proposed in Dartmouth society in 1956. Nowadays, AI, which is at the forefront of high-tech field in the 21st century, plays an important role in promoting economic development, improving domestic people’s livelihood and protecting national and homeland security. In July 2017, the State Council issued the development plan for the new generation of AI, which mentioned that China would be the world’s major AI innovation center by 2030. The report of the 19th National Congress of the Communist Party of China also clearly proposes to promote the further in-depth integration and development of the Internet, AI and the real economy. At the same time of national policy driven development, many AI enterprises are increasing investment in R & D personnel, R & D funds and other elements, optimizing internal structure and management, so as to promote the high-quality development of AI industry, thus driving economic growth. AI media consulting research report pointed out that in 2018, the market scale of China’s AI industry reached 41.55 billion RMB, with a growth rate of 75%. From 2015 to 2018, the compound average growth rate of China’s AI industry market scale is 54.6%, far exceeding the global average. China’s AI industry is in a rising period. With the deepening of the layout of the industry giants such as science and technology, manufacturing, etc., the industry giants have strengthened and improved their industrial chain layout in AI, and some companies continue to increase the in-depth exploration of vertical fields. The scale of the AI industry will further expand, and AI will involve more fields.
However, the efficiency of capital input-output in AI enterprises has been ignored. Capital input is capital, while output is indicators of operating results and equity. The source of capital includes equity capital and debt capital, and assets are the external form of expression. There are many kinds of assets. Different input combinations determine their different output forms and output quantities.
Literature review
Some scholars use the neoclassical economic growth model to estimate the impact of machine intelligence on the economy, and find that the use of machine intelligence will improve the economic growth rate. On the basis of empirical analysis of the collected transnational and cross industry data, some scholars have come to the conclusion that intelligence and automation plays a significant role in promoting economic development. Zhou [1] understood the relatively narrow sense of AI industry as an industry that launched a series of innovative research and development and application in AI algorithms and related technologies. At the same time, he thinks that the relatively generalized AI industry covers all the fields of intelligent algorithm, symbol representation, data calculation, etc., and its process covers every link of data collection, production, storage, operation and application. Cai and Xue [2] regarded AI as a working mechanism to explore human intelligence, and based on this, he creates an AI machine that has human intelligence and provides intelligent services for human activities. Shu [3] pointed out that the essence of AI is to simulate human intelligence and intelligent behavior, integrating computer science, information theory, cybernetics, psychology and other disciplines as an interdisciplinary field. Cai and Yue [4] proposed that the AI industry mainly focuses on the key technologies of AI such as deep learning, machine learning and robotics, covering products and services in different fields such as visual and auditory recognition, intelligent manufacturing, smart city and driverless driving. It is an economic activity integrating production, R&D and sales.
Evaluation model design
Sample selection and index design
According to the WIND database AI index (931071. CSI), select its constituent stocks, and consider the following factors when screening: 1) the listed companies with negative input or output indicators, which are not comparable with other companies, will impact the evaluation result and be eliminated; 2) according to the requirements of DEA model for samples and indicators, try to choose that the number of samples is equal to or exceeds the input and output Double the total number of indicators. In this paper, 48 listed companies of AI in China are selected as samples, that is, decision unit (DMU) in the model.
Referring to the research of Xu and Dai [5], combined with the actual situation of sample data, this paper selects undistributed profit (y1), net profit (y2) and surplus reserve (y3) to reflect the capital output indicators of AI enterprises; selects monetary capital (x1), accounts receivable (x2), inventory (x3) and fixed assets (x4) to reflect the capital input indicators of AI enterprises. See Table 1 for details:
Capital input and output indicators of China’s AI enterprises
Capital input and output indicators of China’s AI enterprises
Index data is from WIND dauase, in which monetary capital, accounts receivable, inventory and fixed assets are from corresponding accounts of sample balance sheet in 2017 and 2018, and undistributed profit, net profit and surplus reserve are from corresponding accounts of sample profit statement in 2017 and 2018, as shown in Table 2.
Original data of capital input and output of AI enterprises (unit: 100 million RMB)
Data envelopment analysis (DEA) is a kind of efficiency evaluation method, which was established by Charles and Cooper in 1978. It is a quantitative analysis method to evaluate the relative effectiveness of comparable similar evaluation objects by using linear programming method according to multiple input indicators and multiple output indicators. DEA analysis has the following three characteristics: The first is the objectivity of evaluation index weight selection. Because the weight of each index determined by DEA is not the weight under the priority sense, but the weight of input and output of each decision-making unit is taken as the variable for evaluation from the perspective of the most favorable decision-making unit; secondly, DEA method carries out comprehensive analysis and processing on the scattered evaluation indexes, and uses the data from the perspective of the whole, thus avoiding the one-sided analysis of the index processing; thirdly, DEA can get the quantitative index of the comprehensive efficiency of each decision-making unit without determining the explicit expression relationship of input-output, then determine the effective decision-making unit, and analyze the causes of the effective decision-making unit, so as to further adjust the positive direction and degree of the input scale of the decision-making unit. Based on the above characteristics and unique advantages of DEA analysis, DEA is recognized as an effective tool to study efficiency problems.
The first DEA model proposed by Charnes, Cooper and Rhodes is C2R model, which is mainly used to analyze and evaluate the relative effectiveness of decision-making units. The second DEA model is called BC2 model, which was given by banker, Charnes and Cooper in 1984. It is mainly used to determine the decision unit whose corresponding point is located on the production function image. The evaluation method is based on the comparison of relative effectiveness, and takes the weight vector of each input and output of DMU as the variable without defining the specific logic function relationship between each input and output factor, which effectively avoids the objective evaluation caused by the subjective decision of the weight. Therefore, choosing BC2 model has better pertinence and applicability.
The method hypothesizes that if there has a decision-making unit (DMU), each DMU in value creation has m kinds of input and s types of output. The value creation input vector of the jth decision-making unit (DMUj) is x
j
= (x1j, x2j, …, x
mj
)
T
> 0. The value creation output vector of the jth decision-making unit (DMUj) is y
j
= (y1j, y2j, …, y
sj
)
T
> 0. And their weight vectors respectively are v = (v1, v2, …, v
m
)
T
and u = (u1, u2, …, u
s
)
T
. The efficiency rating index of the jth decision-making units (DMUj) is:
Suppose the optimal solution for the dual programming
Having a non-Archimedean infinitesimal ɛ BC2 model of the dual form (variable returns to scale) is [6]:
Imported the original data of input indicators and output indicators of 48 Chinese AI enterprises into the software DEAP2.1 and set the parameters of the guidance file. The decision-making unit is 48, the time is one year, the number of output indicators is 3, the number of input indicators is 4, and the output leading type is 1 = output ordered, 1 = VRS, 0 = DEA (multi-stage). Based on BCC model, the operation results of DEA of Chinese AI enterprises are shown in Table 3.
Evaluation results of capital input-output efficiency of AI enterprises
Evaluation results of capital input-output efficiency of AI enterprises
According to Table 3 below, the average comprehensive efficiency of capital input-output of 48 Chinese AI enterprises is 0.68, the average pure technical efficiency is 0.73, and the average scale efficiency is 0.90. The results of DEA operation show that the capital input-output efficiency of 16 Chinese AI enterprises is DEA effective, and the other is non-DEA effective. The detailed analysis is as follows:
16 enterprises, including Teamax Smart City Technology, DHC Software, INESA Intelligent Tech, 360 Security Technology and Qu Mei Home Furnishings Group etc, have achieved DEA efficiency in capital input and output. The comprehensive efficiency of capital input-output of 23 enterprises is higher than the average, and the other 25 enterprises are lower than the average, and the minimum value is 0.11 (Beijing BDStar Navigation).
Among the 32 enterprises without the DEA efficient performance evaluation, 5 have a pure technical efficiency of 1.00, but the scale efficiency is insufficient, such as Unisplendour (0.82), Zhejiang Dahua Technology (0.89), NavInfo (0.98), Hangzhou Hikvision Digital Technology (0.97) and Toyou Feiji Electronics (0.63). The pure technical efficiency and scale efficiency of 27 enterprises such as Digital China Information Service, Inspur Electronic Information Industry and Dawning Information Industry are less than 1.00. It shows that five enterprises such as Unisplendour can gradually achieve the improvement of capital input-output efficiency by expanding the enterprise scale and improving the scale efficiency, while 27 enterprises such as Digital China Information Service need to improve both the pure technical efficiency and the scale efficiency to further improve the capital input-output efficiency.
In the total sample, 9 enterprises are in the state of increasing returns to scale, 23 enterprises are in the state of decreasing returns to scale and 16 enterprises are in the state of not changing returns to scale, accounting for 18.75%, 47.92% and 33.33% of the total sample respectively. The increase and decrease of scale compensation shows that more than half of the enterprises are in the state of scale decline, which may be attributed to the fact that in recent years, the state has vigorously encouraged the development of AI industry, the strength of relevant preferential policies is relatively large, the market share has accelerated growth, and the enterprises have seized the opportunity to continuously expand the model, give full play to the scale advantage, and quickly seize the market.
Table 4 shows the relaxation variables of capital input and output of AI enterprises. For enterprises with better management and more output, efforts should be made to reduce investment, reduce costs and give full play to capital input-output efficiency; on the contrary, other enterprises only need to consolidate the existing foundation and input level.
Comprehensive attribute value and final scores of initial evaluation indices of ecological responsibility of resource-oriented enterprises
Improvement strategy of capital input and output of AI enterprises
From the strategy of capital input-output improvement of AI enterprises, it can be found that there are many areas to be improved in each non DEA effective index of capital input-output of AI enterprises in China. Taking BDStar Navigation as an example, the projection points of its capital input-output efficiency on the production front are (26.17, 9.81, 3.29) and (14.34, 10.26, 5.51, 6.11). Therefore, in the capital input-output efficiency of it, the undistributed profit should be adjusted by 2.289 billion RMB, the net profit should be adjusted by 858 million RMB, and the surplus reserve should be adjusted by 289 million RMB. When DEA is effective in the current period, the receivables of 15 million RMB can be reduced. As its business application is still in the process of promotion and the market and products are constantly developed, the company’s profit and balance in the early stage are lower than expected. When the market scale is expanded and the products are mature, its capital input-output efficiency will be greatly improved. In addition, the output indicators of Hikvision are the highest level of the industry, i.e. the undistributed profit reaches 19.479 billion RMB, the net profit reaches 10.38 billion RMB, and the surplus reserve reaches 3.972 billion RMB. According to the improvement strategy, the projection points of its capital input-output efficiency on the production front are (19479103.80, 39.72) and (215.10156.62, 53.33, 40.53), that is, the pure technical efficiency of Hikvision has been optimized, but the scale efficiency has not been optimized, and the scale reward is decreasing, which may be related to the rapid expansion of the company in recent years, the implementation of diversification and overseas development strategy, the resource integration and scale advantage are not solid synergy.
Based on the above empirical analysis of the evaluation of capital input-output efficiency of China’s AI enterprises, this paper draws the following conclusions: firstly, the comprehensive efficiency of AI enterprises is not high, and 2/3 of the sample enterprises’ comprehensive efficiency does not reach DEA efficiency, which indicates that there is capital waste in the AI industry, which needs to be improved. Secondly, the gap between the comprehensive efficiency of AI enterprises is very large, the highest is 1, the lowest is only 0.11, which shows that the development level and strength of enterprises in the industry show great differences, and the competition is extremely fierce. Thirdly, the average value of pure technical efficiency of AI enterprises is only 0.73, which is the main reason for the low comprehensive efficiency of the industry, and is closely related to the high-tech requirements and innovative types of the industry, which also shows that AI enterprises should focus on independent innovation, constantly explore new products and models, strengthen the research and development of high-end products, and vigorously improve the pure technical efficiency, so as to realize the improvement of overall income. Fourthly, the average level of scale efficiency of AI enterprises is 0.90, showing a high level, but nearly half of the enterprises are in the stage of diminishing returns to scale, which shows that in the stage of rapid expansion, the capital investment is increased, and there is no effective integration, but it is easy to cause the waste of resources, thus reducing returns. Finally, from the perspective of enterprise monetary capital and accounts receivable, the average value is 2.207 billion RMB and 2.159 billion RMB respectively, which is higher than inventory (1.362 billion RMB) and fixed assets (1.001 billion RMB), indicating that the enterprise has sufficient capital reserve, and should increase product research and development, realize product innovation, and prevent the risk of bad debts of accounts receivable.