This paper walks through mathematical evolution of modern portfolio theory and multi-factor models and advances with a General Intelligent Portfolio Theory and underlying applications in stock markets. Following up the earlier form of the Intelligent Portfolio Theory, the new generalization extends in 3 dimensions: 1) three forms of intelligent portfolios – multi-asset multi-strategy, multi-strategy multi-asset and multi-trader; 2) strength investing with momentum rotation as an engine driving dynamic re-selection of assets or strategies or traders; 3) sector rotation in stock markets as a main form of strength investing and as a paradigm shift from diversification in portfolio theory. Applications in Chinese stock markets and international index futures are demonstrated with nontrivial performance achieved through testing on historical data.

Along with the continuous development of capital markets and intelligent finance technologies, quantitative investment is entering into the most critical and challenging area – fundamental quantitative investment. So far, quantitative investment has been focused on automation of technical analysis and trading, while fundamental investment has been large discretionary. This paper provides an overview of quantitative investment and fundamental investment towards a fundamental quantitative investment theory and technical system based on multi-factor models. We start with reviewing relevant literature on modern financial quantitative investment and fundamental investment. Then we cover the theoretical basis and development of multi-factor models and their applications for stock selection, involving linear and non-linear relationships, machine learning, deep learning with neural networks, random forests, and Support Vector Machines (SVMs). We explore the frontiers of fundamental quantitative investment and shed light on the future research prospects.

The statistics for a window of an engineering observation is called running window quantity. Running intra-covariance is a special case of running inter-covariance, when two variables are the same. The inter-covariance minus the geometric average of respected intra-covariances is called the pure inter-covariance. It gives you a cleaner association analysis compared with the running Pearson analysis. The normalization of covariance to standard deviation is called the Pearson correlation coefficient. The covariance for the region above or below the average is called semi-covariance (upper or down). Here we present a pure inter running semi-covariance, an accurate ReLU (Rectified Linear Unit) way of measuring the inter-non-linear correlation between variables excluding the intra-non-linear components. Our framework is applied to successfully analyze the association between war factors and the gold response. The result of our analyses of the 12 years after the 2007 Mortgage crisis on the war equipment companies' stock versus the gold suggests that stocks from different regions have a slightly different impact on the gold value that reflects the overall peaceful economic prosperities.

After nearly three decades of a hard journey, China's capital market has more and more clearly demonstrated the right value of value investing. A-share market participants - retail investors and institutions - are in urgent need of a value investing theory in line with China's national conditions. We realize that China's value investing system must be the joint value investing of China and the world. This paper proposes a value investing theory and quantitative realizing technology system with China as the main body and taking both China and the world conditions into account. The main contents include: 1) under the framework of big data, using the credit risk analysis for filtering out stocks with mediocre or poor credit; 2) multi-factor models of quantitative investment for selection of value and growth stocks; 3) deep learning financial market prediction model for capturing dynamic margin of safety and profit opportunities; 4) deep intelligent portfolio trading technology for implementing value investing into super intelligent systems of quantitative investment. The characteristics and innovations of the theory are: expanding the big data holographic credit risk analysis for Chinese enterprises to value investing analysis; developing comprehensive multi-factor models for selecting value and growth stocks into portfolios; developing big data-driven deep learning financial market prediction models; innovating and developing deep intelligent trading strategies and systems.

The expectation of an engineering observation random quantity is the first order origin moment. Variance is a special case of covariance, when two variables are the same. The variance is the second-order central moment, its root is standard deviation. The normalization of covariance to standard deviation is called Pearson correlation coefficient. The covariance for the region above or below the average is called semi-covariance (upper or down). Here we present semi-covariance, an accurate ReLU (Rectified Linear Unit) way of measuring the non-linear correlation between variables. Our framework is applied to successfully analyze the association between alternative factors and the poll response. The result of our analyses of the 2020 USA presidential election suggest that stock, pandemic, funding, culture, and mental health have different impacts on presidential candidates: Biden vs Trump. The voters care about the economy (stock and funding situations), pandemic impacted voter’s culture fairness income, and voters’ stress leading to mental issue. That’s why we picked above five factors. Whoever win more number of strong correlations is predicted as the winner.

Recently AI technologies, especially Deep Neural Network (DNN), have been widely used in the financial industry, such as stock price and movement prediction. In order to develop an AI-based solution for the global financial market, daily-based DNN model training for each stock is required to need collaboration among a large scale of entities. However, it is usually challenging due to data privacy, the cost of AI computing, and the lack of motivation to share information. This research proposes a novel Blockchain based platform, which utilizes the decentralized network, federated learning, and master-node to tackle these issues. The decentralized computing framework of federated learning, along with transfer learning, is applied to meet the data privacy requirements. Furthermore, the proposed federated learning platform with collaborative training is built on a decentralized AI computing cloud, which is highly affordable compared to centralized AI clouds. The master-node of Blockchain technology is further employed to enable the scalable global financial service, and effective rewards are applied to incentivize information sharing as well. We have applied the proposed Blockchain based platform to the stock prediction global service, which demonstrates the platform is practical and useful.

This paper presents a specific form of the Intelligent Portfolio Theory for trading in commodity futures markets. The theory is based on the recognition of the reality that any single trading strategy is bounded in rationality, so it is unable to remain profitable consistently over time; thus an intelligent portfolio consists of a multi-market portfolio of which capital allocation on each market is managed by a multi-strategy portfolio. In Chinese commodity futures markets, 3 futures are selected; and 2 trading strategies are developed and applied on each of the selected futures. This specific intelligent portfolio trading system is tested on historical intraday data, and the result exhibits superior performance with the least maximum drawdown and the biggest reward-to-risk ratio.

This paper presents a special form of the general Intelligent Portfolio Theory for arbitrage in commodity futures markets. The theory is based on the recognition of the reality that any single arbitrage strategy on any futures pair is bounded in rationality, so it is unable to keep exhibiting desirable performance consistently over time; thus an intelligent portfolio for arbitrage in futures markets consists of a number of selected futures pairs of which each pair is arbitraged via one or multiple dynamic strategies. Here integration test and correlation are used for selecting futures pairs, and Kalman filtering is used to model the trend of the price spread for each futures pair. In Chinese commodity futures markets, 32 futures are examined, from which up to 14 futures pairs are selected. This intelligent portfolio arbitrage system is tested on historical intraday data, and the result exhibits desirable performance with low maximum drawdown and high reward-to-risk ratio in comparison with individual futures pair arbitrages.

This paper proposes a general formalism for representation, inference and learning with general hybrid Bayesian networks in which continuous and discrete variables may appear anywhere in a directed acyclic graph. The formalism fuzzifies a hybrid Bayesian network into two alternative forms: the first form replaces each continuous variable in the given directed acyclic graph (DAG) by a partner discrete variable and adds a directed link from the partner discrete variable to the continuous one. The mapping between two variables is not crisp quantization but is approximated (fuzzified) by a conditional Gaussian (CG) distribution. The CG model is equivalent to a fuzzy set but no fuzzy logic formalism is employed. The conditional distribution of a discrete variable given its discrete parents is still assumed to be multinomial as in discrete Bayesian networks. The second form only replaces each continuous variable whose descendants include discrete variables by a partner discrete variable and adds a directed link from that partner discrete variable to the continuous one. The dependence between the partner discrete variable and the original continuous variable is approximated by a CG distribution, but the dependence between a continuous variable and its continuous and discrete parents is approximated by a conditional Gaussian regression (CGR) distribution. Obviously, the second form is a finer approximation, but restricted to CGR models, and requires more complicated inference and learning algorithms. This results in two general approximate representations of a general hybrid Bayesian networks, which are called here the fuzzy Bayesian network (FBN) form-I and form-II. For the two forms of FBN, general exact inference algorithms exists, which are extensions of the junction tree inference algorithm for discrete Bayesian networks. Learning fuzzy Bayesian networks from data is different from learning purely discrete Bayesian networks because not only all the newly converted discrete variables are latent in the data, but also the number of discrete states for each of these variables and the CG or CGR distribution of each continuous variable given its partner discrete parents or both continuous and discrete parents have to be determined.