The new generation of information technology revolution represented by 5G, big data, artificial intelligence, etc. is profoundly affecting various fields, aspects and links of the current economic and social operations. Technological changes have become a profound motivation to promote the change of macro-control concepts and method innovation. . Exploring and establishing a set of fast, effective, adaptive, intelligent, and accurate economic monitoring and forecasting methods is not only an important support for further effectively improving the level of macroeconomic regulation and control, but also a historical mission of the government to accelerate compliance with the changing times. It is also able to achieve high-quality development for our country, Creating a high-quality life and promoting high-level openness provide a stronger driving force.
The Party Central Committee with Comrade Xi Jinping as the core attaches great importance to the important work of combining big data intelligence with economic monitoring and forecasting, and clearly stated at the Third Plenary Session of the 19th Central Committee: "To strengthen economic monitoring, forecasting and early warning capabilities, comprehensively use big data, "Technical means such as cloud computing have enhanced the forward-looking, targeted, and synergistic nature of macro-control." The role of next-generation information technology in economic monitoring, forecasting and early warning has been raised to a new level. Guided by actual application requirements, this article attempts to analyze the challenges faced by current economic monitoring and forecasting methods, and explains the overall ideas, key issues, and countermeasures of a completely data-driven economic monitoring and forecasting method in order to provide useful ideas.
I. Four Challenges Facing Traditional Economic Monitoring and Forecasting Methods in Complex Economic Formations
(1) Difficulty in timeliness: it is difficult to judge the latest situation in real time based on historical statistics
The current economic monitoring and forecasting methods are mostly based on historical statistical data for analysis or forecasting future data changes, but this method is often subject to the timeliness of statistical data. Traditional statistical data is mainly used to carry out economic censuses, sampling surveys, key surveys, typical surveys and other methods, and uses the data reporting method to collect and calculate corresponding macroeconomic indicators. The most obvious drawback of this method is that it has a strong time lag. Sex. When responding to major emergencies and high-frequency economic shocks, traditional economic monitoring and forecasting methods are likely to lead to government policy-making departments' hindsight and prone to serious consequences such as decision errors.
(II) Difficult to subdivide: the limitations of statistical samples easily lead to insufficient analysis granularity
The number of statistical samples and the cost of sample acquisition are often inversely proportional. As the number of statistical samples increases, the time cost, labor costs, and capital costs in the statistical process will increase correspondingly. Therefore, the coverage of the statistical samples has certain limitations. Sex. Because traditional economic monitoring and forecasting methods rely heavily on statistical data samples, once the samples are biased, on the one hand, it is easy to cause inaccurate analysis and prediction results, on the other hand, it is difficult to provide effective support for economic decision-making in subdivided industries and emerging industries.
(3) Difficult objective: factors such as human misreporting may cause the result to be less objective
The generation of traditional statistical data requires a large amount of manual collection, processing, reporting, and aggregation. Not only is the level of human participation deep, the time period consumed is long, but the statistical process is also complicated. Because humans have participated in too many links in the whole process, whether it is subjective considerations or objective errors, it is very easy for data to be late, underreported, concealed, and misreported. At this time, the authenticity of statistical samples often exists. Hidden dangers of deviations. If this kind of data with deviations is used for economic monitoring and prediction, the scope of errors will be further expanded.
(IV) Accuracy is difficult: Economic monitoring and forecasting has not been closely integrated with the most advanced technology
Economist-led economic monitoring and forecasting models have relatively mature economic theories as the support. They mainly include structured econometric forecasting models, unstructured time series forecasting models, and dynamic stochastic general equilibrium forecasting models, which can be better explained. The key influencing factors of economic and social operation. However, in today's world, deep-seated contradictions in global development are prominent, protectionist and unilateralist trends of thought have risen, the multilateral trading system has been impacted, and China's economy has also shifted from a high-speed growth stage to a high-quality development stage. The overall development environment of the global economy faces many risks and uncertainties. In the face of this intricate change, the assumptions of traditional measurement methods often do not conform to the current economic reality. At the same time, traditional empirical theories often deviate from the actual situation, and traditional economic monitoring and forecasting models are slightly weak. At present , the deep learning technologies represented by BERT, XLNet, ROBERTa, etc. have shown extraordinary prediction capabilities and practical value in many fields, but they have not been deeply integrated with economic monitoring and prediction.
Overall thoughts and key issues of a completely data-driven economic monitoring and forecasting method
As discussed earlier, analyzing or forecasting future changes in data based on historical statistics will present difficulties such as timeliness, segmentation, objectiveness, and accuracy. Under such circumstances, the effective use of micro data and new technological methods to improve economic monitoring and forecasting methods has become a hot issue of common concern for government policy makers and academia. A group of top scientists represented by Professor Varian, an authority on microeconomics, pointed out that big data methods represented by technologies such as machine learning, deep learning, and complex networks are changing the research paradigm of economics. A number of top journals around the world have also launched special topics on "monitoring and forecasting". Through a number of research and practice results, they have interpreted the latest research progress and future directions of new technologies on economic and social monitoring and forecasting in many aspects. A large number of theoretical methods and practical experience have been To some extent, the possibility of a completely data-driven economic monitoring and forecasting method has been demonstrated.
(I) Overall thinking of a completely data-driven economic monitoring and forecasting method
The overall idea of a completely data-driven economic monitoring and forecasting method is shown in Figure 1, which can be summarized as: "N + 1 + 3", that is, to aggregate N types of multi-source data, break through a key link, and precipitate three types of data assets.
Aggregate N kinds of multi-source data. It is necessary to introduce sustainable and stable access to fine-grained data resources, including but not limited to complaints, referee documents, corporate associations, real estate prices, employment and recruitment, satellite lighting, patent papers, search engines, road congestion, news public opinion Relevant data such as successful bidding, website operation, etc.
Break through a key link. On the basis of fully absorbing the theoretical foundations and model methods of economic monitoring and forecasting at home and abroad, a large-scale application of information technology such as machine learning, deep learning, complex networks, and natural language processing is required. On the one hand, it can reduce the degree of control of artificial assumptions and human experience. On the other hand, a large number of data features can be extracted from multi-source fine-grained data, and the combination of data features is continuously tried to automatically collide with the results of the monitoring and prediction targets, and the data features and corresponding algorithms are continuously optimized and adjusted according to the monitoring and prediction results. Machine automation selects an optimal combination to achieve economic accurate monitoring and prediction.
Depositing three data assets. It is necessary to take "the number of things as the guideline" as the guiding ideology, continuously deposit process data in various stages of economic monitoring and forecasting, use technical means to deepen the breadth and depth of data extraction, and use the data feature database, algorithm model library, and forecast that are deposited in practice. The index library is continuously forcing the technical methods to iteratively upgrade to speed up the gap between economic monitoring and forecasting and new technologies.
Figure 1 Overall idea of a fully data-driven economic monitoring and forecasting method
(II) Key Issues of Fully Data-Driven Economic Monitoring and Forecasting Methods
The completely data-driven economic monitoring and forecasting methods need to focus on solving the following three types of key problems: First, the problem of data supply. How to obtain fragmented micro data scattered all over the society in real time, quickly, accurately, and stably is an important support and guarantee of a completely data-driven economic monitoring and forecasting method, and it has become one of the key issues at present. The second problem is feature engineering. Data and features determine the upper limits of machine learning and deep learning, and models and algorithms only approach this upper limit as much as possible. Therefore, how to provide as many feature extraction ideas as possible for fully data-driven economic monitoring and prediction methods is the second key issue. The third is the core technical issues. Whether the core technology of interdisciplinary monitoring and prediction can be broken is the key issue and the third difficult point, mainly including the ability to use natural language processing, complex networks, deep learning and other related technologies in feature extraction to map high-dimensional space features through mapping or The transformation method is converted to low-dimensional space dimensionality reduction technology, covering machine learning and deep learning technology of classification prediction and numerical prediction, and data visualization technology to simplify the results.
3. Suggestions on accelerating the use of fully data-driven economic monitoring and forecasting methods
(I) Speed up the construction of a real-time collection and aggregation mechanism for multi-source data
Economic society is a complex system composed of numerous entities such as government agencies, enterprises, individuals, vehicles, goods, media, and various social organizations. Therefore, it is necessary to strengthen the entity objects such as people, enterprises, vehicles, things, things, and land. Unified collection and collation of desensitized data resources, mainly including business flow data, global commodity transaction data, terminal vehicle positioning data, high-scoring satellite remote sensing data, various types of Internet data, and other related data that are closely related to the production and operation process of the enterprise Resources.
(II) Focus on promoting economic monitoring and forecasting in key areas
Fully consider the major strategic needs of countries such as coordinated regional development, military-civilian integration, rejuvenating the country through science and education, strengthening the country with talents, innovation-driven development, rural rejuvenation, and sustainable development. Focus on the key links of a completely data-driven economic monitoring and forecasting method and accelerate the construction of key areas. Feature engineering system, real-time dynamic monitoring system, prediction and early warning perception system, and comprehensive research and judgment support system.
(3) Prospective research on key technologies for economic monitoring and forecasting
Organically cooperate with government agencies, university think tanks, scientific research institutes, enterprises and institutions, financial institutions, etc. to conduct forward-looking interdisciplinary research on key technologies for economic monitoring and forecasting, and strengthen the regression comparison and correlation of multi-source heterogeneous data characteristics with traditional statistical indicators. Analysis, to form a joint force of "government, industry, research, and research funding" oriented to economic monitoring and forecasting .
(4) Establishing a working mechanism for linkage of economic monitoring and forecasting
Relevant government departments at all levels should encourage governments at all levels to actively participate in the demonstration work of fully data-driven economic monitoring and forecasting applications. All parties share the application results, actively publicize and demonstrate the effectiveness of economic monitoring and forecasting applications at all levels of government departments, and provide a mechanism for economic monitoring and forecasting at the national level. Build the foundation quickly.