In the era of digital economy, economic activities based on Internet, mobile Internet and artificial intelligence are generating massive data, which have promoted economic growth in return. Big data have various sources and forms, including structured and unstructured data. Unstructured data can be text, images, audio and video, and new structured data can be functional data, interval data, symbolic data, etc. Most Big data has a huge volume, and some is Tall Big data, whose dimensionality of potential explanatory variables is larger than its sample size. The rise of Big data together with machine learning, a main computer-based automatic analytic tool for Big data, has profound implications on statistical science. Based on the characteristics of Big data and the nature of machine learning, the paper discusses the challenges and opportunities brought by Big data and machine learning to statistical modeling and inference, including sample inference, sufficiency principle, data reduction, variable selection, model specification, out-of-sample prediction and causality. We also explore the theory and methodology foundation of machine learning and its integration with statistics.

We selectively review some literature on prediction in the presence of big data. Issues of data based approach versus causal approach, micro versus macro modeling, homogeneity versus heterogeneity, model uncertainty versus sampling errors, constant parameter versus time-varying parameter modeling, model evaluation and cross-validation as well as aggregation, etc. are considered.

This paper aims to introduce the philosophy, theories, fundamental content systems, models, methods and tools of modern econometrics based on its development history. We first review the classical assumptions of the linear regression model and discuss the historical development of modern econometrics by various relaxations of the classical assumptions to further illustrate the modern theoretical system and fundamental contents. We also discuss the challenges and opportunities for econometrics in the Big Data era and point out some important directions for the future development of econometrics.

This paper provides an overall view on the SAR model, which generalizes the autoregessive time series model to a spatial setting. It is the most popular model in spatial econometrics with broad applications in empirical economics as it captures interactions and spilled over effects across economic agents. We first provide some economic justification of such a model in an complete information static game setting, of which observed outcomes are Nash equilibria. Comparative statics analysis in economics provides economic implications on direct and indirect effects and multiplier effect on outcomes. The traditional ML estimation and its extension in terms of QML estimation are discussed. Recent developments on concavity of its log likelihood function are established, and alternative estimation methods, GMM and GEL, are presented. The construction of best GMM estimation with linear-quadratic moments is feasible. The GEL approach on estimation and testing for the SAR model can be robust against unknown heteroskedasticity.

This survey paper highlights some recent developments in estimating treatment effects for panel data. First, this paper begins with a brief introduction of the basic model setup in modern econometric analysis of program or economic policy evaluation for panel data. Second, the primary attention goes to the focus on estimating both the average and quantile treatment effects for panel data. Finally, it concludes the paper by addressing theoretically, methodologically and empirically some possible future research directions for young scholars in econometrics and statistics, particularly, some interesting and challenging research topics related to a combination of machine learning and casual inference for panel data.

There is an increasing trend towards combining machine learning methods with traditional econometric methodologies. Starting from comparing features and internal relations of two mainstream causal inference frameworks, this paper proposes that causal inference can be significantly improved with the introducing of machine learning methods in two ways, one is sample matching and one is counterfactual prediction. Firstly, machine learning techniques can enhance matching qualities by pairing samples directly or improving the accuracies of propensity score predictions. This can make the matched samples more similar to samples collected from randomized controlled trials. Secondly, machine learning techniques can improve the accuracies of counterfactual predictions by modeling complex relations, using cross-validation, and adopting regularization. This paper then introduces the theoretical foundations of combining machine learning techniques and causal inferences by reviewing four specific methods: Matching, regression discontinuity, difference-in-difference, and synthetic control method. At the meantime, several application cases are provided in each method section for researchers in applied econometrics as references.

From the perspective of economics, we introduce the economic interpretations and applications of some basic concepts, ideas, methods and tools in probability and statistics. Examples include subjective probability, cumulative distribution function, stochastic dominance, quartile, expectation, Jensen's inequality, mean and variance, law of large numbers, variance of sample mean converging to zero, independence, martingale difference sequence, statistical significance, goodness of fit and model specification, spectral analysis of time series, machine learning and random sets. Economics applications consist of rational expectations, depicting income inequality, portfolio, efficient market hypothesis, capital asset pricing model, financial derivatives pricing, risk management, model risk, measuring economic causal relationships, identifying economic cycles, and macroeconomic interval management. These examples and applications are helpful in understanding the importance and usage of probability and statistics in economic research and analysis.

It is urgent to develop systematic theories and methods to support the scientific research of innovation management and innovation policy. At the same time, the scientometric theories and methods of analyzing the upstream scientific and technological output of the innovation process fail to meet the needs of comprehensively analyzing the whole innovation process and systematically supporting the research of innovation management and innovation policy. This situation inevitably leads to a more comprehensive interdisciplinary "Innovation Metrology (Innovametrics)". Innovametrics is an emerging discipline that takes the entire innovation system as the research object, orients to the occurrence and development of innovation, and comprehensively analyzes the innovation system. Innovametrics constructs the theoretical and methodological system for analyzing the innovation process from the perspective of innovation system, so as to realize the systematic diagnosis and analysis of the innovation process. Based on the simultaneous development of innovation activity analyses and innovation models, this paper classifies Innovametrics into innovation input metrics, innovation output metrics, innovation profit metrics, innovation transformation metrics and innovation system metrics from five aspects: Input (I), output (O), profit (P), transformation (T) and system (S). Typical research problems and analysis techniques of Innovametrics are summarized from the perspectives of structural, development and dynamic problems. Finally, combined with the practice of innovation management in China, the urgent scientific problems of Innovametrics are prospected. The continuous enrichment of innovation statistics and surveys will inevitably promote the vigorous development of Innovametrics, and the development of Innovametrics will also make innovation management and innovation policy more scientific. The "I-O-P-T-S" five-dimensional Innovametrics system constructed in this paper not only provides a classification system for the analysis of innovation activities for the first time, but also provides a systematic whole-process analysis perspective for the design of research problems in the field of innovation.

Characteristics selection approach is quite critical to multi-factor model for stock selection. Based on data from the Chinese A-share stock market, this paper uses Group LASSO to select characteristics and to nonparametrically estimate the effect of selected factors on future returns. We find that, although a few selected factors are the same as traditional characteristics selection models, many (e.g., current ratio, de-trended turnover, price-to-earnings ratio) are unique. In addition, we use the selected characteristics to predict 1-month-ahead returns, and construct a portfolio going long 20 stocks with the highest predicted returns. We show that, compared with portfolios generated by traditional models, portfolios based on Group LASSO and non-parametric estimation approaches perform better, with higher abnormal return and greater Sharp ratio. Furthermore, selected factors using Group LASSO and non-parametric estimation approaches are quite different between the Chinese and US stock markets. For example, momentum (or reversal), and volatility which are selected factors in the US stock market are not related to future stock return in the Chinese stock market; price-to-earnings ratio and current ratio, which are selected characteristics in the Chinese stock market, are not significant in the US stock market.

The purpose of this paper is to systematically state how to use the Gibbs sampling method as a tool, based on the inference principle of big data association feature factors to conduct the extraction of related risk characteristics in the context of financial derivatives. Specifically, using stochastic search method based on Gibbs sampling algorithm under the framework of Markov chain Monte Carlo (MCMC), incorporating the "odd ratio" as criteria, taking FOF, commodity futures as examples, this report systematically discusses how to extract related risk factors for financial derivatives based on big data (including traditional and non-structure data) approach as shown by using so-called stochastic research method. Furthermore, we like to point out the way to identify the risk factors based on big data method established in this paper is quite now and should be useful for the financial service in the practice of Fintech.

In recent years, natural language processing (NLP) technology has been widely used to study the emotional polarity of unstructured text data such as financial news, financial commentary and social media, and the emotional polarity of these unstructured text data are utilized as proxy variables of investor sentiment to predict the volatility of financial market. Based on behavioral finance theory, an exchange rates forecasting with sentiment mining of online foreign exchange news is proposed in this dissertation by means of NLP and deep learning. This approach uses mutual information theory to construct the first sentiment lexicon in the field of foreign exchange. On the basis of sentiment lexicon of foreign exchange, the sentiment polarity of foreign exchange news is calculated by combining the basic lexicon constructed in this dissertation. The study shows that there is a Granger causality and long-term co-integration relationship between the sentiment polarity of foreign exchange news and USD/CNY exchange rate. Additionally, this study incorporates the sentiment polarity data of foreign exchange news and other financial data into deep learning approach. The empirical results show that our proposed approach has a significant effect on short-, medium-, and long-term volatility forecasting of USD/CNY exchange rate.

Developing effective systematic risk measures plays an important role in preventing systematic risk. Based on detailed information of 16 Chinese commercial banks and daily returns of China Securities Index (CSI) 300 from Jan-2011 to Dec-2019, we estimate the systematic risk measure of SRISK for the Chinese banks, which is the expected shortfall of a bank conditional on a prolonged market decline, by means of Monte Carlo simulations. The SRISK is applied to reveal risk ranking of banks, uncover risk contagion among industries, and predict macroeconomic indicators. Our results highlight that: 1) SRISK delivers robust rankings of systematically important banks; 2) Despite the fact that some commercial banks are in the same category, they have heterogeneous effects on different industries. State-owned banks are contagious to the entire industry; 3) SRISK can be used to predict PPI and RPI in short term. It is also capable to predict CPI in 2~3 months. Our results indicate that SRISK not only can be incorporated into the risk management system to strengthen prudential supervision of the banking industry, but also satisfies the requirements of macroprudential supervision due to its bank-industry sensitivity.

The socio-economic field is one of the main occasions for data science applications. How to grasp the disciplinary pattern and focus of data science in this field is a basic problem in formulating and implementing discipline development strategies. Based on the practical needs of China's economic statistics and the important points in the "HMYW 2019 Statistics Report", this paper discusses the disciplinary pattern of data science applications in the socio-economy. We propose that "data processing methods", "ambiguity uncertainty phenomenon" and "problem-driven pattern" are the three focuses of the application of data science in the socio-economy in the era of big data, analyzing their respective importance and key points. Finally, we discuss the future development direction and strategic adjustment of data science.

For the study of text, the mathematical principle of representation learning is firstly introduced, in which distributed representation makes the representation of text information richer and more effective, especially for the ultra-high-dimensional text big data. Taking the famous BERT Transformers model as an example, its high-dimensional representation vector (768 dimensions) codes grammar and semantic knowledge. This information can be reconstructed using a decoder. The main contributions of this paper are in two aspects. The first aspect is to introduce statistical methods to evaluate the representation ability of this BERT model; the second aspect is to use this model to deal with the difficulty of identifying the ambiguity of Chinese sentence segmentation. In the first aspect, we have two findings. One is that BERT model has sufficient representation capability. Even for the text data of 100, 000 samples, dimensional reduction of vector representation space can be obtained to a large extent, which indicates that BERT model has reserved sufficient representation vector space to contain various complex languages. We find that different layers of representation vectors of BERT model of deep learning represent different levels of information of language knowledge. The first layer mainly represents the information of words, while the deeper the depth is, the closer the representation vector is to the whole language knowledge of text. Second, it is found that semantically similar sentences are also in the similar domain in BERT vector space, which indicates that the whole BERT vector representation space adaptively arranges similar language organization in the similar subspace. In the second aspect, we cleverly make use of the MASK mechanism of the BERT model, which allows some words to be hidden at will in the input sentences of the model. However, the model can adaptively predict the representation vector of the part that is deliberately hidden. By comparing the representation vector after correct and wrong segmentation, we were able to correctly identify which molecular approach was correct with an average accuracy of 66.875%.

CPI and industrial added value are basic and important indicators of national economic accounting, and have important practical significance for the analysis of the macroeconomic situation. This paper makes an out-of-sample prediction of China's CPI and industrial added value based on a factor model that better characterizes China's economic characteristics, and compares and analyzes with traditional prediction models. The empirical results support that the factor model has better prediction capabilities than traditional methods.

Stock index price change reflects the trend of the equity market, is an important financial market index, has been the focus of academic circles. This paper taking the CSI300 price index as the research object and the data from January 2006 to March 2019 is selected, use H-LSTM model to predict the stock index price. The result show that: Using batch prediction, point-by-point pushback, and controlling the rise and fall, as long as increasing the number of network input variables and reducing the time width of the forecast can increase the fitting degree of the LSTM model to the index price; The hybrid information extractor is better than the principal component analysis, the sparse Autoencoder and the t-SNE, it can effectively extract the characteristics of the CSI300.

Dissatisfaction in feeling state is crucial factors causing decision-making bias and economic loss among middle-aged and elderly. Based on the tracking data of the 2015 China Health and Retirement Longitudinal Study (CHARLS), this paper empirically analyzes the impact of happiness on the fraud risk in middle-aged and elderly. The results show that there is a significant negative correlation between the happiness of middle-aged and elderly people and the risk of being cheated. Besides, we find the result is still robust after controlling endogeneity and other methodological alternatives. Moreover, this paper provide enough evidence to support the hypothesis that happiness could prevent middle-aged and elderly from bearing fraud risk by increasing their self-control capacity. Against the background of frequent frauds among the middle-aged and elderly, this paper helps to understand the role of mental health in their economic decision-making, offering policy makers suggestions to establish a fraud prevention system for middle-aged and elderly.

The changing international situation and global market environment are accompanied by a series of "Black Swan" and "Gray Rhino" events. Estimating the shock effect of significant emergencies and predicting price change points have always been among the hot and challenging issues the academic circle faces. This paper developed the GSI-BN research framework to analyze the shock effect of significant emergencies on the oil market and predict the oil price trend when different events occur. This paper explores the impact of emergencies and forecasts the crude oil price by developing the GSI-BN research framework. First, we construct the network public opinion attention index of emergencies and determine the time windows of different emergencies based on the Google Search Index (GSI). Secondly, the impact mechanism of emergencies is simplified to the topological network diagram based on the Bayesian network (BN). The emergencies are subdivided and the conditional probability behind them is mined to predict the occurrence probability of the emergencies. Finally, we analyze and forecast the price caused by emergencies in different scenarios. The results show that when the monthly average growth rates of supply and demand are high, supply and demand are still in a state of balance, and the price will not rise or fall sharply. Therefore, the probability of low price is the largest. When the supply shock is significant and the demand is growing at a normal level, the probability of a medium-low price is the largest. On the demand side, when the financial crisis occurs, the probability of absolute growth of crude oil consumption in the next month was highest in the medium growth range; On the supply side, the simultaneous occurrence of two or three emergencies is a small probability event. In addition, the global oil demand will rise as OPEC is committed to reducing production. Therefore, the impact of hurricanes on prices is smaller than in previous years. The effect of the financial crisis on the crude oil market is comprehensive, and the demand shock caused by the financial crisis is only one of the factors. Both the war and the OPEC meeting were short-lived supply shocks and primarily reflected market expectations, and they may not cause a supply side shock. Therefore, there will not be a significant price difference when the impact is transmitted to the oil price. This paper provides a new perspective and method for studying the shock effect of emergencies on the crude oil market and the change point of oil price.

Forecasting stock prices has been regarded as one of the most challenging tasks, because stock prices are determined by many factors, including macroeconomic indicators, fundamental factors and investors sentiment etc. Traditional time series models forecast stock returns only based on daily, weekly or monthly historical observations. Recently, the fast development of the financial markets and data collection technology make high-frequency data more and more available. As high-frequency data may contain some extra information compared to low-frequency data, it may be able to improve stock prices forecasting. In this paper, we propose a new method based on nonparametric functional data analysis, which is used to extract predictors from high-frequency data. Then we combine the high frequency predictors with traditional time series predictors to be a mixed prediction model for forecasting stock prices. The construction method of predictors is very flexible, without any parametric assumption. For empirical studies, we apply the method to CSI300 index, and find that the new method after incorporating the useful information from high-frequency data performs better than the traditional autoregressive model. Our results provide an empirical evidence that high-frequency data could help improve stock prices forecasting.

With the increasing global attention to carbon dioxide emissions, the carbon market price has become more and more important. Accurate and reliable carbon market price forecasts can not only provide a better reference for the government's macro-control to achieve the "Chinese carbon peak and carbon neutralization" goal, but also help enterprises can more effectively manage the risks brought by carbon emissions. This article uses empirical mode decomposition (EMD), convolutional neural networks (CNN) and long short-term memory networks (LSTM) to predict carbon emissions trading prices to propose different EMD-CNNs-LSTM combination strategy, and discussed the effect of the CNN-LSTM combination strategy, divide it into serial strategy and parallel strategy. This article uses the Guangdong, Shanghai, Hubei, and Chongqing carbon market price data to systematically compare the EMD-CNN-LSTM combination strategy with single prediction model and combination prediction model. The prediction effect of EMD-CNN-LSTM verifies the accuracy and robustness of the four combined strategy models of EMD-CNN-LSTM, demonstrated that parallel strategy forecasting is better for the universality of carbon market price series.

With the development of information technology, social media have gradually become an important channel for people to obtain information, share views, and vent feelings, and exerted a profound impact on information dissemination and sentiment contagion in the capital market, giving birth to a large body of literature focusing on social media and capital market. The extant literature has confirmed social media's information content and role in influencing the behaviors of investors, companies as well as other information intermediaries. This paper uses the method of bibliometrics to sort out and analyze the literature in this area. The main results are as follows: 1) This research area is in a stage of vigorous development, and the amount of publications exhibits an increasing trend over the years; 2) The core effects of journals for research in this area have not fully emerged; 3) The research team in this field has basically formed; 4) The research topic in this area can be divided into two parts: Social media and stock trading and social media and corporate governance. Based on these findings, this paper summarizes the possible research directions as well as issues worthy of attention in the hope of providing a valuable reference for scholars who are interested in this area.

This study investigates the recent development of financial research. We first summarize the up-to-date progress of research on asset pricing, corporate finance, and the economic development and finance market. Then we analyze emerging trends and challenges and show the following frontiers of financial research, including new monetary theory to accommodate financial crisis and digital currency, sustainable finance, finance safety, and climate finance. Among these topics, there are significant opportunities for China's finance research, especially in the area of preventing financial systemic risk, expanding and monetary theory, and developing inclusive finance.

The main goal of macro prudential policies is to maintain financial stability. This paper proposes adopting the macro-econometric policy evaluation method under the Rubin causal effect framework to evaluate the impact of China's macro prudential policies on financial stability during the sample period 2007--2020. First, the paper constructs a macro prudential policy index to quantitatively measure the intensity of China's macro prudential policies. Second, the paper uses the systemic financial risk index, termed as SRISK to measure China's systemic financial risk. Finally, the paper evaluates the macro prudential policies' effects on the systemic financial risk, cross-sectoral contagion of systemic financial risk and important intermediate variables in the credit channel. Our empirical findings indicate that loose macro prudential policies can increase the risks of intermediate variables in the credit channel, and the risks lead to a significant rise in SRISK of house sector, but for the SRISK of financial and manufacturing sectors, the cumulative effects in 24 periods are not significant. However, in addition to a significant rise in commercial banks' capital adequacy ratio growth, tight macro prudential policies have no significant effects on the other intermediate variables in the credit channel, and further have no obvious effects on SRISK of financial, house and manufacturing sectors. Based on the conclusions, we suggest that systemic risk indicators should be further researched to provide more comprehensive and systematic targets for macro prudential authorities. Moreover, the transmission channel of macro prudential policies on financial stability should be improved to enhance the efficiency of regulation. Finally, more attentions should be paid to the cross-sectoral contagion of systemic financial risk so as to prevent systemic financial risk from a systemic perspective.

Central bank communication is an important narrative text that receives a lot of attention from the market, and how to effectively extract key information from the high-dimensional text is a scientific problem to be studied in depth. In this paper, we apply the Sentiment Extraction via Screening and Topic Modeling method proposed by Ke et al. (2019) to measure central bank communication, which has the advantages of simplicity, transparency and replicability. Considering the characteristics of Chinese texts and the multi-instrument framework of China's monetary policy, we select the change values of several actual monetary policy interventions as supervised variables and then construct a central bank communication index, and forecast future actual monetary policy interventions based on generalized monetary policy rules. The results show that textual information on central bank communications helps to provide additional forecasting power. Compared with the indexes constructed by the existing literature based on text analysis methods such as keywords, supervised dictionaries and LDA methods, the index constructed in our paper has better forecasting power, especially with superior performance in long-term forecasting. We verify the effectiveness of central bank communication in guiding expectations from a predictive perspective, and provides feasible solutions for extracting textual information based on different target indicators.

This paper proposes proportion of gainers (POG) to reflect the heterogeneity of social program. This parameter reflects the proportion of different subgroups who benefit from the participation of the social program in the audience, which can be used to reflecting the universality of the policy benefiting the people. Under the general framework of causal analysis of latent variables, this paper studies the nonparametric identification of POG, and proposes a feasible estimation method. Through numerical simulation, it is found that the proposed estimator has good small sample properties. We use this method to estimate the effect of the New Rural Pension Scheme (NRPS) on the labor supply, family income and family consumption of rural residents. And we find that the NRPS has obvious heterogeneity. TheNRPS has a greater role in promoting individuals with high labor supply, high family income and high family consumption, but a weaker role in promoting individuals with low labor supply, low family income and low family consumption. This shows that in order to comprehensively and accurately evaluate a livelihood policy, we should pay attention to the heterogeneity of the policy.

Large models, exemplified by ChatGPT, represent a significant breakthrough in general generative artificial intelligence technology. Their far-reaching implications extend into diverse facets of human production, lifestyle, and cognitive processes, prompting a transformative paradigm shift in the realm of economic research. Originating from the convergence of big data and artificial intelligence, these large models introduce a novel approach to systemic analysis, particularly adept at scrutinizing intricate human economic and social systems. We first discuss the fundamental characteristics and development paradigms of ChatGPT and large models, focusing on how these models effectively tackle the methodological challenges posed by the "curse of dimensionality". We then delve into how ChatGPT and large models will influence the paradigm of economic research. This includes a shift from the assumption of the rational economic man to an AI-driven "human-machine hybrid" economic agent, from the isolated economic individual to the socio-economic individual whose behaviors are measurable, from the separation of macroeconomics and microeconomics to their integration, from the separation of qualitative and quantitative analysis to their unification, and from the long-dominant "small-model" paradigm to a "large-model" paradigm in economic research. We also cover the increasing significance of computer algorithms as a prominent research paradigm and method in economics. Finally, we point out the limitations inherent in artificial intelligence technologies, including large models, when employed as a research method in economics and the broader social sciences.

Accurate short-term load forecasting is one of the key technologies to ensure the safe and stable operation of new power systems. However, residential load forecasting faces the difficulties of huge number of users, high load heterogeneity, high volatility and high randomness. With the increase of user types and data, the complexity of the model will increase significantly, making it difficult to meet the requirements of load forecasting in new power systems. Therefore, this paper develops a structured long- and short-term neural network model based on prediction-oriented autoencoders, which can accurately forecast the short-term load of all types of users through a single model. Compared with similar models, the prediction accuracy of the 13 combined models proposed in this paper is improved by 7.16%~11.59%, and it is also of great referential significance for the unified prediction of complex high-frequency time series of highly heterogeneous subjects in non-electricity fields.

Dynamic double spatial autoregressive (DDSAR) model is one of popular models in spatial econometric analysis, and is widely used in applications for its generality. This paper provides a new tool in a DDSAR model-impluse response analysis. The new tool captures the average change of dependent variable due to one-unit change of explanatory variable and the spatial structure existing in disturbances and dependent variables. Following the classical studies (LeSage and Pace (2009)), we decompose the effect into direct, indirect and total effects, and define the respective dynamic values over time and the accumulated ones. The paper provides the estimation method and inferential theory on these values. Monte Carlo simulations confirm our theoretical results and show good finite-sample performance of the estimators. We apply the new econometric tool to the regional economic development of China, and particularly investigate the nexus of human capital and GDP per capita. We find that the impulse response of indirect effects is positive in the short term and negative over the long term, implying that the relationship of Chinese regional economies are win-win in the short term and competitive over the long term. Policy suggestions are made.

This paper studies the dynamic effects of monetary and fiscal policy shocks on China's fixed asset investment and total retail sales of consumer goods. The main conclusions are as follows: 1) The results of linear impulse responses of policy shocks shows that only the loan interest rate shock has a significant negative impact on fixed asset investment in the short term, while the effects of monetary supply policy and fiscal spending policy shocks on domestic investment and consumption are statistically insignificant; 2) The results of state dependent impulse responses of policy shocks show that loan interest rate shock has a short-term significant negative effect on fixed asset investment and retail sales of social consumer goods in periods of expansion but the impact is not significant during recessions; 3) money supply and government spending policy have no significant impact on real investment and consumption, no matter whether in expansions or in recessions. Based on these findings, we suggest that large-scale policy stimulus should not be implemented in a recession.

Since the reform and opening up, especially during 2008-2019, China's M2/GDP has been rising for a long time without serious persisting inflation and China's velocity of money is declining continually, which is called "the monetary puzzle in China". In order to explain this phenomenon, based on China's important economic characteristics and more than 30 types of actual data, this paper expands the model structure in the literature on real estate bubble, establishes and calibrates a general equilibrium model. After accounting for two kinds of money demands with leverage effects and two money supply drivers, real estate investment and local government debt, the calibrated model maintains a low CPI growth rate and explains 42.27% of the 12-year M2/GDP rise, and 49.70% of the decline in the velocity of money. The "counterfactual" analysis shows that if the money supply basically matches the nominal GDP growth rate according to the requirement of "14th Five-Year Plan of China", it can solve the "the monetary puzzle in China" and enhance the high-quality economic development in China. The main contribution of this paper is to provide an analytical framework that can quantify the impact of real estate investment, local government debt, and other factors on the rise in M2/GDP and the decline in the velocity of money.

Magnet effect is the phenomenon that the price accelerates toward the price limits or circuit breaker when price approaches them. We design a quasi-direct comparison method based on counterfactual simulation to test the effect near price limit events in China's stock markets. We first construct the theoretical conditional probability distribution of price movement under the counterfactual condition that there isn't price limit policy, and compare the theoretical and practical conditional probabilities. The results show that magnet effect exists in China's stock markets, and is stronger near lower price limit than near upper price limit. Further, we find that magnet effect is stronger in harder-to-value stocks and more active-trading stocks. Magnet effect is stronger when a price limit event occurs at the middle of the trading day than what at the opening period of the trading day. One possible reason of magnet effect might be institutional investors' demand for liquidity.

Based on the stock price return data of 86 listed financial institutions in China, this paper uses CoVaR, cross-section VaR, absorption ratio, Granger causality index, and information spillover index to measure systemic financial risk. After ddetailed inspecting lead-lag relationship between these indexes and their abilities to predict the macro economy, the results indicate that: CoVaR, absorption ratio, Granger causality index and information Spillover index all have a certain ability to predict the macro economy in advance, but the cross-section VaR does not have the forecasting power; Granger causality index has a certain lead over other indicators, while the absorption ratio and Spillover index are relatively lagging. In summary, the Granger causality index is a good systemic financial risk measure and has a robust economic forecasting ability.

Accurately predicting the crude oil prices is vital for governors to make policies and essential for market participants to make investment decisions. We propose a hybrid multi-step-ahead forecasting model that integrates the secondary decomposition algorithm which combines variational modal decomposition (VMD) and integrated empirical modal decomposition (EEMD), differential evolution (DE) and extreme learning machine (ELM), namely, VMD-RES.-EEMD-DE-ELM, for more accurate crude oil price forecasting in this paper. To illustrate the superiority of the proposed model, the sample data of Brent and West Texas Intermediate (WTI) are used to validate the performance of the proposed model. The empirical results confirm that the proposed model achieves better performance compared to several other benchmark models in terms of forecasting accuracy and stability.

This paper investigates the relationship between idiosyncratic volatility and stock returns in A-share market from a perspective of transmission mechanism. A typical idiosyncratic volatility puzzle is discovered in A-share market but unlike in developed countries, its main cause isn't value return but the rotation of short-term speculation and value return. Speculation driven by retail trading exerts a significantly positive influence on stock returns in the future. This influence weakens to show the effect of value return only when idiosyncratic volatility begins to decrease, which is an important signal of switching point in rotation.

The value added of digital economy is an important statistical indicator reflecting the development scale of digital economy and its contribution to the overall economy. So far, the concept, scope and measurement methods of digital economy are still in the process of exploration, there's no unified opinions and consistent measurement methods. Thus, there are significant differences between the results of digital economy value added that calculated by different academic institutions and scholars. This paper systematically combs the definition of digital economy form narrow scope and wide scope; summarizes three method of digital economy value added: Production method in GDP accounting, method based on growth accounting framework and econometric method; compares the results of the value added of digital economy of the United States and China, which have relative importance in the world. Finally, this paper summarizes the challenges, and puts forward corresponding suggestions. Try to improve the calculation method of value added of digital economy and the comparability of calculation results, meanwhile, attempt to provide reference to formulate digital economy policies and promote the high-quality development of digital economy.

There are considerable nonstationary time series in economics, finance, climate science and related areas. In last two decades or so, in order to improve theoretical research in these disciplines, asymptotic theory on nonstationary time series has captured close attention and well developed; on the other hand, classical series estimation often requires the values of variables considered fall into a bounded compact interval that in some circumstance suppresses the development and application of the method in nonparametric context, especially in the present of nonstationary time series. In order to break through the bottleneck of the conventional sieve method, the authors and their coauthors use orthogonal series expansion to achieve some theoretical results and their applications, in particular in nonparametric and nonstationary time series. These studies lay a foundation for the use of the series estimation in economics, finance, climate science and related disciplines.

Since OpenAI launched its artificial intelligence generative content (AIGC) product — ChatGPT on November 2022, the whole world has turned upside down. The AIGC mainly come two main streams: Large language models (LLMs) and diffusion models. New application and research publish daily in an accelerated way. In this paper, we first raise a serious question over LLMs: Does intelligent ability generated from LLMs really owns general artificial intelligence (AGI, artificial general intelligence) ability like ordinary people's intelligence ability doing things?In this paper, I first make an important hypothesis: As a closed system, through a large language model (LLM) has been designed to represent, store human's huge knowledge and intelligence's ability and behavior, equipped with the highest value standard that the model must align to human value, but the LLM model doesn't demonstrate its AGI ability. However, as an opened system, once we input some formatted text with implicit human's knowledge and intelligence, then we suddenly find that output of the LLM model show natures of certain human's intelligence and behavior. The formatted input text is called a prompt. The higher intelligent prompt is, the better intelligent output of the model will be. In other words, the LLM models own some kind of AGI ability conditioned on prompt.Economics research and other social science research such as politics, history, and linguistics include the most complex social forms and the deepest human minds, so in this paper, we try to explore whether AGI-like general AI for large language models LLMs is a fact or an illusion by summarizing the latest research results of other researchers? For this model's AGI-like general AI capabilities, we summarize the latest research results of these research scholars, including issues such as IQ levels of large language models, AIGC industrial economics, computational social science research under AIGC, business decision making, and virtual AIGC economist paradigm research in economics and other social sciences.

The real growth rate of gross domestic product (GDP) is an important indicator to measure the state of the economy. However, as it is released quarterly with a time lag, it fails to meet the timely economic analysis demand. In this paper, the mixed frequency dynamic factor model (MF-DFM) is used to nowcast quarterly GDP year-on-year growth rate based on timely large-scale monthly economic data, which improves the timeliness of economic analysis. In order to enhance the efficiency of utilizing a large number of available candidate economic variables and avoid the subjectivity of indicator selection in the factor model, this paper proposes a indicator selection method for MF-DFM with large-scale data, which uses the mean square prediction error of the binary dynamic single factor model as the basis for indicator selection. This method is applicable to selecting effective indicators amongst data with quarterly and monthly frequencies, missing values and jagged edges. The empirical analysis results indicate that compared with the traditional time series prediction models and the commonly used mixed frequency models, the MF-DFM based on screening variables by the binary model has a higher accuracy in predicting quarterly GDP growth rate, both for the stability period before COVID-19 and the recovery period after COVID-19. Moreover, the prediction for monthly GDP growth rate provided by this method has a high synchronization with the macroeconomic consistency index, which is conducive to improving the timeliness of economic analysis. This paper provides a new approach for real-time economic monitoring, prediction, and early warning based on indicator selection with the large-scale data.

Due to the characteristics of non-stationary prices and high noise, it is obvious that the prediction of high-frequency prices in financial markets is difficult. Unlike other forecasting methods, the forecasting method based on the hybrid neural network model does not only predict the next price, but turns the price forecast in the future into a forecast of future trends and duration, that is, the prediction of multi-point becomes a prediction of two variables, and this prediction method is more efficient. This paper uses the mixed neural network model to make price predictions for the smooth price data of the Shanghai and Shenzhen 300 stock index futures. First, based on 5-minute time series data of the Shanghai and Shenzhen 300 stock index futures, this paper uses a hybrid neural network model for trend prediction and compares it with LSTM and CNN models. Then use the mixed neural network rolling forecast to design an investment strategy: If the predicted trend is up, go long in the future for a period of time, otherwise go short. This article backtests the price data for 2016, 2017, and 2018, and compares the performance of investment strategies with buy-and-hold strategies. The results show that after deducting fees, the strategy based on hybrid neural networks is better. Finally, we performed a stability test and tested the usability of the model from a practical perspective.