AI in FinTech: A Research Agenda

07/10/2020 ∙ by Longbing Cao, et al. ∙ 0

Smart FinTech has emerged as a new area that synthesizes and transforms AI and finance, and broadly data science, machine learning, economics, etc. Smart FinTech also transforms and drives new economic and financial businesses, services and systems, and plays an increasingly important role in economy, technology and society transformation. This article presents a highly summarized research overview of smart FinTech, including FinTech businesses and challenges, various FinTech-associated data and repositories, FinTech-driven business decision and optimization, areas in smart FinTech, and research methods and techniques for smart FinTech.



There are no comments yet.


This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

I Introduction

In recent years, finance has become increasingly interactive with the new-generation artificial intelligence (AI) techniques and data science (DS)

[1, 2, 3, 4, 5]. In particular, FinTech (or Fintech, i.e., financial technology in full) [6, 7] is at the epicentre of synthesizing, innovating and transforming financial services, economy, technology, media, communication, and society broadly driven by AIDS techniques [8]. Here, AIDS broadly refers to both classic AI areas

including logic, planning, knowledge representation, modeling, autonomous systems, multiagent systems, expert system (ES), decision support system (DSS), simulation, pattern recognition, image processing, and natural language processing (NLP) and

modern AI and DS areas

such as representation learning, machine learning, optimization, statistical modeling, mathematical modeling, data analytics, knowledge discovery, complexity science, computational intelligence, event and behavior analysis, social media/network analysis, and more recently deep learning and cognitive computing. In contrast, we broadly refer

finance to areas including capital market, trading, banking, insurance, leading/loan, investment, wealth management, risk management, marketing, compliance and regulation, payment, contract, auditing, accounting, financial infrastructure, blockchain, financial operations, financial services, financial security, and financial ethics. In addition, economics and finance (EcoFin for short) are increasingly synergized with each other and with AIDS.

AIDS is the keystone enabler of the new-generation of EcoFin and FinTech [9, 10, 11, 12]. The new-generation AIDS is reshaping or even redefining the concepts, objectives, content and tasks of EcoFin and FinTech. AIDS essentially and comprehensively transforms the way and effect that modern economic and financial (economic-financial) businesses operate, transact, interact and collaborate with their participants (incl. consumers, markets, and regulators) and environments. AIDS nurtures new economic-financial mechanisms, models, products, services, and many tangible and intangible opportunities. As a result, AIDS not only strengthens the efficiency, cost-effectiveness, customer experience, risk mitigation, regulation, and security of existing economic-financial systems but also innovates unprecedented and more intelligent, efficient, convenient, personalized, explainable, secure and proactive economic-financial products and services. In this paper, we call the synthetic product of AIDS and EcoFin as smart FinTech.

Figure 1 presents a four-dimensional, systematic and interactive landscape of the synthesis between AIDS and EcoFin to form smart FinTech. The landscape connects the main EcoFin businesses (bottom) to the EcoFin data and repositories (left), the broad-based AIDS techniques (right), and the EcoFin business objectives (top). In the rest of this paper, we will discuss and summarize the main business domains, technical areas, and techniques associated with smart FinTech in terms of the FinTech businesses and challenges, FinTech business objectives, FinTech data, areas in smart FinTech, and AIDS techniques for smart FinTech. We will conclude with opportunities for future smart FinTech developments.

Fig. 1: A Four-dimensional Research Landscape of the Synthesis between AIDS and EcoFin toward Fostering Smart FinTech. Shapes in different colors represent distinct dimensions in the synthetic landscape. Each dimension initiates its interactions and communications with the three other dimensions through its same-colored, directed connectors and channels. Each channel carries business, problem, data (incl. information and knowledge), intelligence, and technology, etc. from one end to another. The four dimensions interact with each other to address specific economic-financial business problems of underlying businesses by particular AIDS techniques on the corresponding data.

Ii FinTech Businesses and Challenges

The FinTech business areas that build on AIDS techniques involve almost all aspects of an economic-financial system and its environment and broadly all EcoFin businesses [1, 12]

. Here, we highlight the following major business areas in smart FinTech: (1) economic-financial innovations: e.g., new mechanisms and products; (2) economic-financial markets: including products and services; (3) economic-financial participants: including individual and retail investors, institutions and regulators; (4) economic-financial behaviors: e.g., investor activities and company announcements; (5) economic-financial events: e.g., company merger and financial crisis; (6) economic-financial services: e.g., banking, insurance, lending, financing, and crowdfunding services; (7) economic-financial mechanisms: e.g., market mechanisms, business models, and derivative pricing; (8) economic-financial systems: e.g., a company’s finance; (9) economic-financial infrastructure: including fundamental supporting systems and blockchain; (10) economic-financial pricing: including the valuation of underlying and derivative assets in capital markets; (11) economic-financial trading: including trading, investment, and execution; (12) economic-financial payment: e.g., online, mobile and contactless payment; (13) economic-financial valuation: e.g., the estimation of property value, credit and intangible assets; (14) economic-financial marketing: including campaign, and customer care; (15) economic-financial relationship management: e.g., stakeholder relations, and business partnership; (16) economic-financial resource management: including management of human, material, informational and intangible assets; (17) economic-financial operations: e.g., the processes and services for supporting financial innovation, design and production; (18) economic-financial compliance and regulation: e.g., enforcing operational orders and business rules by authorities; (19) economic-financial crisis, risk and security: e.g., investment risk, systemic risk, and cybersecurity; and (20) economic-financial ethics: including social and ethical issues and privacy.

Major FinTech challenges associated with the above FinTech business areas as well as their associated EcoFin businesses, problems, data and objectives in Fig. 1

can be categorized into (1) innovation challenges: e.g., AIDS techniques for inventing novel, efficient, intelligent and sustainable mechanisms, products, services and platforms; (2) business complexities: such as AIDS techniques for representing, learning and managing the intricate working mechanisms, structures, interactions, relations, hierarchy, scale, dynamics, anomaly, uncertainty, emergence and exceptions associated with a market, a product or a participant; (3) organizational and operational complexities: such as AIDS techniques for understanding and managing the diversity and personalization of individuals and departmental teams, the departmental and institutional coherence and consensus, and the inconsistent and volatile efficiency and performance; (4) human and social complexities: such as AIDS techniques for modeling and managing the diversity and inconsistency of participant’s cognitive, emotional and technical capabilities and performance, and for enabling effective communications, cooperation and collaboration within a department and between stakeholders; (5) environmental complexities: such as AIDS techniques for modeling and managing the interactions with contextual and environmental factors and systems and their influence on a target business system and problem; (6) regional and global challenges: such as understanding and managing the relations between an economy entity and its financial systems with the related regional and global counterparts and stakeholders and their influence on a target problems; (7) data complexities: such as extracting, representing, analyzing and managing data quality issues, misinformation and complicated data characteristics, e.g., uncertainty and extreme dimensionality, sparsity and skewness; (8) dynamic complexities: such as modeling, predicting and managing evolving but nonstationary behaviors, events and activities of individual and block markets, products, services and participants; and (9) integrative complexities: e.g., systematically modeling and managing the various aspects of the above complexities that are often tightly and loosely coupled with each other in an underlying economic-financial system.

Iii The Synthetic Areas of Smart FinTech

Figure 2 summarizes the synergy between major AIDS techniques and finance areas to generate various synthetic smart FinTech areas. Our discussion focuses on AIDS techniques for transforming the following smart FinTech areas [12, 7]: smart banking, smart insurance, smart lending, smart marketing, smart payment, smart regulation, smart risk, smart security, smart trading, smart wealth, and smart blockchain, in addition to the foundations for smart FinTech. Each of these primary areas is further explained in terms of specific problems and tasks where AIDS techniques can play distinct and significant roles [3].

Fig. 2: The Synthesis between AI/DS Techniques and Financial Businesses to Form Smart FinTech. The upper fin refers to broad AI and DS areas and techniques, the lower fin consists of major financial areas, and the ridge includes various synthetic areas of smart FinTech.

The foundations of smart FinTech

refer to the AIDS areas and techniques to enable smart FinTech. Examples are actionable financial intelligence; AIDS for financial infrastructure; analyzing big economic-financial data; analyzing cloud, online and mobile economic-financial services; analyzing environmental, social and governance data; analyzing global economic-financial event and impact; analyzing high-dimensional, sequential and evolving economic-financial data; analyzing multisource, multimodal and heterogeneous economic-financial data; automated economic-financial systems and services; computational intelligence for EcoFin; data mining and knowledge discovery for EcoFin; deep learning and representation for EcoFin; distributed learning, federated learning and transfer learning for EcoFin; intelligent cross-market economic-financial data processing; intelligent economic-financial recommendation; modeling economic incentives; modeling economic mechanisms and social welfare; modeling evolutionary equilibrium of financial markets; modeling economic-financial chaos, uncertainty and change; modeling economic-financial couplings, interactions and dependencies; modeling economic-financial information flow; modeling economic-financial information propagation and influence; modeling financial market microstructure; modeling economic-financial mechanisms, models, products and services; modeling economic-financial networking and interconnections; modeling economic-financial structures and hierarchy; modeling natural, online, socioeconomic, political and cultural factors; personalized economic-financial recommendation; reinforcement learning for EcoFin; and statistical and mathematical learning for EcoFin.

The smart banking are active, personalized, automated, trusted, robust, secure, risk-averse, and fragility-averse banking businesses and services. AIDS research for enabling smart banking includes AIDS techniques for detecting, analyzing and managing bank network risk and risk contagion; AIDS for modeling, detecting and managing banking risk and fraud; AIDS for conducting credit analysis and improving pricing; AIDS for optimizing, validating and risk-managing credit loans; AIDS for creating, mining, securing, risk-managing and optimizing cryptocurrencies; AIDS for creating, securing, evaluating, risk-managing and optimizing digital currencies; AIDS for personalizing, automating, validating, securing, risk-managing and optimizing Internet/online banking; AIDS for enabling, personalizing, automating, securing, risk-managing and optimizing open banking; AIDS for evaluating, securing and risk-managing shadow banking; and AIDS for making smarter and more automated, personalized, mobile and engaging banking services; etc.

The smart insurance enables insurance businesses and services that are proactive, tailored, trustful, resilient, secure and risk-averse. The AIDS research for enabling smart insurance includes AIDS for tailoring individual, business and commercial insurance products and services; AIDS for insurance fraud detection; AIDS for enabling personalized recommendation of insurance products and services; AIDS for evaluating, analyzing, detecting, managing and optimizing insurance risk and compliance; AIDS for evaluating, automating, detecting and optimizing insurance security; AIDS for creating novel insurance models, products and services; AIDS for evaluating, automating and improving insurance business operations; and AIDS for integrating, recommending and optimizing multi-policy, multi-product and cross-selling insurance products and services; etc.

The smart lending supports lending businesses and services that are efficient, risk-averse, personalized, context-oriented, predictive, robust and secure. AIDS research directions for smart leading cover problems and tasks such as AIDS for addressing various aspects and problems of blockchain; AIDS for automating crowdfunding, e.g., campaign design and strategy optimization; AIDS for creating and optimizing distributed ledger technologies; AIDS for fundraising, e.g., creating personalized recommendations, efficient fundraising models, and intelligent services; AIDS for analyzing, detecting, managing and mitigating lending risk and security; AIDS for enabling smart, efficient and low-risk peer-to-peer lending; and AIDS for creating, validating and optimizing smart contracts; etc.

The smart marketing

offers marketing activities and services that are cost-effective, relevant, proactive, personalized, positive, sequential and context-oriented. AIDS-enabled smart marketing involves many research areas, such as AIDS for conducting consumer and client sentiment analysis; understanding consumer and client opinion and intention; AIDS for evaluating and optimizing customer relationship management; AIDS for validating, enhancing, synergizing and optimizing econometrics; AIDS for evaluating, risk-managing, personalizing and optimizing stakeholder relationship; AIDS for making, evaluating and optimizing financial recommendations; and AIDS for enabling, automating, evaluating, securing, risk-managing and optimizing supply chain finance; etc.

The smart payment includes online, mobile, wifi and contactless payment methods and services that are secure, risk-averse, fast and convenient. Smart payment becomes increasing important in digital society and economy. AIDS-enabled smart payment research include areas such as AIDS for validating, automating, securing and risk-managing contactless payment; AIDS for enabling, securing, risk-managing and optimizing e-payment; AIDS for automating, securing, risk-managing and optimizing Internet and online payment; AIDS for automating, securing, risk-managing and optimizing mobile payment; AIDS for modeling, detecting, analyzing and mitigating payment risk; and AIDS for validating, detecting, analyzing and mitigating payment security; etc.

The smart regulation supports automated or human-machine-cooperative, risk-sensitive, proactive, systematic, dynamic and evidence-based regulation and compliance operations, governance and risk control. AIDS-empowered smart regulation consists of many research topics, such as AIDS for enhancing corporate finance transparency; AIDS for evaluating and optimizing corporate governance and regulation; AIDS for making and optimizing cross-market regulation; AIDS for creating and optimizing digital currency regulation; AIDS for detecting, analyzing, risk-managing and managing financial crime; AIDS for verifying, automating, securing and optimizing financial digital authentication; AIDS for enabling, validating, verifying, detecting, risk-managing and optimizing financial digital identification; AIDS for quantifying, validating, monitoring, detecting, analyzing and mitigating financial fragility, crisis and stability; AIDS for recognizing, detecting, analyzing and mitigating financial fraud; AIDS for enabling smarter and more efficient and personalized financial market regulation, design and policy implication; AIDS for automating, evaluating and optimizing financial operations; AIDS for quantifying, evaluating and managing information asymmetry and transparency; AIDS for enabling, automating and optimizing international regulation; AIDS for validating and improving market legitimacy; AIDS for quantifying, verifying, detecting, analyzing and improving market social justice; and AIDS for quantifying, evaluating, automating and improving marketplace trust and coordination; etc.

The smart risk and security offers whole-of-time, business and customer risk and security management, which are predictive, systematic, evidence-based, and dynamic. AIDS can play critical roles with many research tasks customizable for specific economic-financial businesses, products and services. Typical AIDS-enabled smart risk and security tasks include AIDS for automating, evaluating and optimizing anti-money laundering; AIDS for monitoring, detecting, categorizing, factorizing, predicting and intervening client financial security, retail investor risk, financial systematic risk, financial institutional risk, financial network risk, and cross-market risk; AIDS for detecting and quantifying financial risk factors and predicting and evaluating financial risk factors and areas; AIDS for quantifying, characterizing, evaluating, predicting and improving consumer confidence and sentiment, market reputation and trust; AIDS for quantifying, analyzing, detecting, predicting and mitigating financial crisis and crisis contagion; AIDS for categorizing, monitoring, detecting, analyzing, predicting, evaluating and managing financial event, market movement, change, exception and emergence and their consequences and impact; AIDS for modeling, detecting, evaluating and managing financial system security, financial network security, and financial instrument security; AIDS for characterizing, quantifying, analyzing, evaluating, predicting and managing financial system vulnerability; etc.

The smart trading supplies predictive, active, dynamic, risk-averse, anti-fragility, and high-utility trading strategies and services. Smart trading involves a very broad and widely explored area. The AIDS-driven smart trading involves many research topics, e.g., AIDS for making and optimizing algorithmic trading, arbitrage trading, high frequency (cross-market) trading, and institutional trading; AIDS for modeling, automating and optimizing digital asset pricing; AIDS for enabling smart e-commerce; AIDS for enabling smart, efficient, personalized and secure Internet finance; AIDS for designing, automating and optimizing novel, secure, smart and personalized financial mechanisms, financial models, financial products, and financial services; AIDS for conducting portfolio analytics and enabling smart portfolio management; AIDS for predictive trading and strategic trading; AIDS for designing, automating and optimizing trading strategies, algorithms and platforms; AIDS for enabling and optimizing socially responsible investment; AIDS for modeling, evaluating and optimizing trading complementarity and substitutionary; and AIDS for characterizing and improving trading incentive and campaign; etc.

The smart wealth becomes increasing topical, with such expected businesses and services for safe, secure, personalized, anti-fragility, and automated wealth management. AIDS-supported smart wealth involves many emergent research areas, including AIDS for making, evaluating and optimizing data monetization; AIDS for valuating, analyzing, evaluating, optimizing and managing digital assets, Internet wealth, public welfare, institutional welfare, and superannuation; AIDS for enabling, automating, risk-managing and optimizing digital financial advising; AIDS for designing and optimizing novel, personalized, secure and healthy financing mechanisms; AIDS for enabling, automating, evaluating and optimizing roboadvising; and AIDS for supporting secure, personalized, low-risk and sustainable venture capital management; etc.

The smart blockchain enables efficient, secure, risk-tolerant, automated or semi-automated, dynamic, and high-performing blockchain infrastructures, computing and services. Both classic and advanced AIDS techniques can play critical roles in enabling smart blockchains. Examples are predicting the price movement of cryptocurrencies, constructing risk-averse cryptocurrency trading strategies and portfolios, detecting risk and assuring smart contracts, efficient mining of bitcoins in distributed systems, supporting proactive and systemic blockchain governance and regulation, and detecting risk, intrusion and fragility in blockchain systems, behaviors and activities, etc.

Iv AI Techniques for Smart FinTech

The AIDS techniques required to achieve the aforementioned smart FinTech objectives and EcoFin businesses are very broad and diversified. Each technique may play distinct roles in addressing respective EcoFin business challenges and data complexities. Here, we group the main AIDS techniques for smart FinTech into the following technical families [13, 14, 9, 15, 16, 17, 18, 19, 20, 3, 21, 22, 23, 24, 25]: (a) Mathematical and statistical modeling,(b) Complex system methods,(c) Classic analysis and learning methods,(d) Computational intelligence methods,(e) Modern AIDS methods,(f) Hybrid AIDS methods.Table I summarizes these AIDS techniques and their applications for smart FinTech.

AIDS areas AIDS methods AIDS techniques Examples of smart FinTech applications
Mathematical modeling Numerical methods Linear and nonlinear equations, least square problem, finite difference methods, dependence modeling, Monte-Carlo simulation Valuation, pricing, portfolio simulation, portfolio optimization, capital budgeting, hedging, price movement prediction, risk modeling, and trend forecasting.
Time-series and signal analysis State space modeling, time-series analysis, spectral analysis, long-memory time-series analysis, nonstationary analysis Price prediction, trend forecasting, market movement prediction, IPO prediction, equity-derivative correlation analysis, change detection, financial crisis analysis, trading strategy discovery, and cross-market analysis.
Statistical learning methods Random walk models, factor models, stochastic volatility models, copula methods, nonparametric methods Market trend forecasting, pricing, valuation, price estimation, VaR forecasting, financial variable dependency modeling, portfolio performance estimate, and cross-market analysis.
Random methods

Random sampling, random walk models, random forest, stochastic theory, fuzzy set theory, quantum mechanics

Abnormal behavior analysis, outlier detection, market event analysis, market movement modeling, influence transition analysis, associated account analysis, crowdsourcing modeling, and marketing modeling.

Complex systems Complexity science Systems theory, complex adaptive systems, chaos theory, random fractal theory System complexity modeling, market simulation, market mechanism design, globalization analysis, crisis contagion, and market information flow.
Game theory Zero-sum game, differential game, combinatorial game, evolutionary game, Bayesian game Interaction modeling, policy simulation and optimization, regional conflict modeling, mechanism testing, coalition formation, and cryptocurrency mechanism testing.
Agent-based modeling Multiagent systems, belief-desire-intention model, reactive model, swarm intelligence, reinforcement learning Testing economic hypotheses, simulating policies, supply/chain relation modeling and optimization, cooperation analysis, self-organization modeling, portfolio optimization, and reinforcement learning.
Network science Linkage analysis, graph methods, power law, small worlds, contagion theory Modeling entity movement, community formation, interactions and linkage, influence and contagion propagation; pool manipulation analysis, and analyzing investor relations.
Classic analytics and learning Pattern mining methods Frequent itemset mining, sequence analysis, combined pattern mining, high-utility pattern mining, tree pattern, network pattern, knot pattern, interactive pattern Trading behavior analysis, abnormal trading analysis, outlier detection, investor relation analysis, customer profiling, high-utility trading pattern analysis, and cross-market trading behavior analysis.
Kernel learning methods Vector space kernel, tree kernel, support vector machine, spectral kernel, Fisher kernel, nonlinear kernel, multi-kernel methods Price and market movement prediction, cross-market time series analysis, financial crisis analysis, crowdfunding estimate, market dependency modeling, and customer profiling and classification.
Event and behavior analysis

Sequence analysis, Markov chain process, high-impact behavior, high-utility behavior, nonoccurring behavior analysis

Financial event analysis, investor behavior analysis, price co-movement prediction, abnormal behavior analysis, market exception and change analysis, market event detection, and group behavior analysis.
Document analysis and NLP Language models, case-based reasoning, statistical language model, Bayesian model, latent Dirichlet allocation, Transformer, BERT Financial event analysis, investor sentiment analysis, company valuation, financial review and auditing analysis, misinformation and rumor analysis, automated question-answering, and keyword searching.
Model-based methods

Probabilistic graphical model, Bayesian networks, expectation-maximization model, clustering, classification, deep neural models

Hypothesis testing, customer clustering, price prediction, trend forecasting, index modeling, event analysis, fraud detection, movement forecasting, valuation, and risk scoring and prediction.
Social media analysis Topic modeling, sentiment analysis, emotional analysis, influence analysis, linkage analysis, interaction learning Social influence analysis, investment linkage analysis, associated account analysis, sentiment analysis, influence modeling, market and price movement, detecting manipulation and insider trading, and understanding company branding and development.
Computational intelligence methods Neural computing methods

Wavelet neural network, genetic neural network, recurrent neural network, deep neural network

Macroeconomic and microeconomic factor correlation analysis, valuation and pricing modeling, relation analysis, sequence modeling, portfolio optimization, and trend and movement prediction.
Evolutionary computing methods

Ant algorithm, genetic programming, self-organizing map, artificial immune system, swarm intelligence, neural-genetic algorithm

New product simulation, financial objective optimization, portfolio optimization, marketing strategy optimization, price and policy testing, market risk analysis, and market performance optimization.
Fuzzy set methods Fuzzy set theory, fuzzy logic, fuzzy neural network, genetic fuzzy logic Modeling market momentum, financial solvency analysis, risk and capital cost modeling, and market uncertainty modeling.
Modern AIDS methods Representation learning Probabilistic model, graph network, network embedding, tree model, neural embedding Representation of stocks, assets, capital markets, portfolios, financial events, behaviors, and financial reports.
Short and informal text analysis Conceptualization, term/tag/phrase similarity learning, dependency parsing, word embedding, deep neural models Text-based trend forecasting of price, market, sentiment and reputation, question/answering modeling and recommendation.
Optimization methods Nonlinear, stochastic and dynamic programming, information theory, Bayesian optimization Optimizing policies, portfolios, trading strategies, VaR, and market performance, cost-benefit optimization.
Reinforcement learning Bellman Equation, actor-critic model, Markov dynamic progress, deep Q-network, adversarial reinforcement learning Simulating and optimizing supply/demand of new assets and services, optimizing portfolios and trading strategies, option valuation optimization.
Deep learning methods Convolutional neural network, attention network, generative adversarial network, autoencoder, deep Bayesian network Market modeling, behavior modeling, trading modeling, risk analysis, price and movement prediction, financial event modeling, cross-market analysis
Hybrid AIDS methods Parallel ensemble Evolutionary neural models, deep Bayesian model, copula graph neural network, combining complexity science and game theory Price and market movement forecasting, multi-aspect risk analysis, macro/microeconomic factor analysis, financial event detection, customer profiling, blockchain and cryptocurrency modeling.
Sequential and hierarchical hybridization Time-series analysis plus classification, macro/microeconomic dependency modeling, deep sequential modeling-based event detection Financial review-based fraud detection, macroeconomic influence on market movement, social media impact on price movement, epidemic evolution, and impact on index.
Cross-disciplinary hybridization Deep multi-time series analysis, copula deep models, autoregression deep model, behavioral economics and finance Psychological factor and irrational market behavior analysis, behavioral finance, misinformation and mispricing on market inefficiency, financial ethics modeling, interpretable financial modeling.
Behavioral economics and finance Prospect theory, nudge theory, natural experiment, experimental economics, behavior informatics, intention learning, next-best action modeling Modeling investment rationality, irrational behaviors, mispricing, market efficiency, mental activities, decision-making process, intention, emotion, and cognitive activities.
TABLE I: AIDS Techniques and Their Representative Applications for Smart FinTech.

Iv-a Mathematical and Statistical Modeling

The family of mathematical and statistical modeling lays the foundation for characterizing, formulating and modeling economic-financial systems and their working mechanisms, problems and solutions. Accordingly, here, we discuss four areas that are widely used for analyzing economic-financial systems and enable smart FinTech.

Numerical methods

are used to build quantitative representation and analysis of economic-financial systems and problems, e.g., for value-at-risk (VaR), option valuation, option pricing, portfolio simulation, portfolio optimisation, hedging, and capital budgeting. Typical methods include linear and nonlinear equations, least squares problems, interpolation, optimization, binomial and trinomial methods, finite difference methods, dependency modeling, financial simulation, Monte-Carlo simulation, random number generators, and econometric models such as term structure modelling and regression.

Time-series and signal analysis

describe, characterize, analyze and forecast the temporal movement, its transition, regression, trend and change by treating economic-financial indicators as time-series and signals. Time-series analysis and signal processing are widely used in EcoFin. For example, the signals of a security price and its derivative price and market index and their relations can be analyzed by multivariate regression and modeling high-frequency trading signals jointly by time-domain sequence analysis and frequency-domain wavelet. Typical methods include state space modeling, linear time-series analysis, nonlinear time-series analysis, time-frequency and time-scale analysis, spectral analysis, Kalman filter, fractional time-series analysis, long-memory time-series analysis, seasonal time-series analysis, transfer function models, multiscale analysis, multivariate analysis, stationary analysis, nonstationary analysis, and outlier analysis.

Statistical learning methods

measure, estimate and learn the uncertainty, randomness, risk, pricing, rating, performance or dependency of economic-financial systems, products and problems in terms of probabilistic theories. Examples are estimating the pricing of options, forecasting the VaR and performance of a portfolio, modeling the sequential trading behaviors of a group of investors or a firm by coupled hidden Markov models, modeling the high-dimensional dependency between multiple time series by a copula method, and measuring the probabilistic relations between financial indicators as a Bayesian network. Typical methods include random walk models, efficient portfolio theories, factor models, Black-Sholes models, Monte-Carlo methods, Delta-Gamma approximation, L

vy processes, stochastic volatility models, copula methods, filtering methods e.g. particle filters, and nonparametric methods such as Bayesian networks and Markov networks.

Random methods characterize, model, simulate and analyze an economic-financial problem in terms of theory of randomness and uncertainty. For example, one can model the evolution of abnormal and exceptional market behaviors, global events, and black swan events by random methods. Typical random methods include random sampling, random walk models, random forest, stochastic theory, fuzzy set theory, and quantum mechanics.

Iv-B Theories of Complex Systems

The theories of complex systems have been widely used in classic AI-driven EcoFin research and applications. They are powerful for understanding, simulating and analyzing the working mechanisms, system characteristics and complexities, and emergence and consequences of economic and financial systems and problems. We introduce four of such methods below.

Complexity science methods model an economic-financial system (e.g., a bitcoin market) as a complex system. Then, complexity science can understand its intrinsic and intricate working mechanisms and system complexities, global economy and its evolution, inter-regional and inter-country relation analysis, migration, crisis contagion, conflict modeling, international trading and information flow. Combined with sociology and systems theory, typical methods include theories of complexities, systems theory, emergence, self-organization, complex adaptive systems, ABM, chaos theory, and random fractal theory.

Game theory methods build mathematical and logic models to design, characterize, simulate and analyze the interactions, conflict, cooperation, communication, coalition, uncertainty, behavioral relations, social norm, reputation, trust, Nash equilibrium, and consensus-building mechanisms and processes in complex economic and financial systems. They can be used to model the conflict between political systems, estimate rational and irrational threats in regional conflict, and model the market mechanisms of blockchain and cryptocurrencies by continuous game theories, to name a few. Typical methods include zero-sum games, continuous games, differential games, combinatorial games, evolutionary games, stochastic games, Bayesian games, strategic-form games, extensive-form games, and the communication, bargaining, cooperation, coalition in collective and cooperative games.

Agent-based modeling and simulation model and simulate the working mechanisms, dynamic formation processes, internal interactions between entities (i.e., agents), and learning and evolutionary processes of economic and financial systems. ABM builds multiagent systems to simulate an economic-financial problem as a complex system, and incorporates such mechanisms as perception, interaction, policy selection, rules, reinforcement learning and optimization to simulate the working mechanisms in such systems. ABM has been widely explored in economics and finance.

Network science characterizes, models, analyzes and predicts the directed and undirected connections and interactions, entity movement, community formation and influence propagation and contagion in economic-financial systems by modeling such systems as complex networks. Typical methods include network linkage analysis, graph methods, scale-free and power law, small worlds, influence diffusion and contagion theories and tools. Such methods can be used to model group investor interactions, relations and communications as a directed network, e.g., by Poisson factorization-based Bayesian models or modeling the interactions between market participants in an economic market.

Iv-C Classic Analytics and Learning Methods

Classic analytics and learning methods have played critical roles in making EcoFin intelligent. This can be achieved by analyzing economic-financial data, discovering the patterns, clusters, classes, trends and outliers in economic-financial systems. We categorize such methods into the following six methods.

Pattern mining methods model and discover patterns and patternable behaviors, structures and relations in economic-financial systems. Examples are identifying arbitrage trading behaviors by mining frequent and high-utility cross-market investment strategies, discovering high-frequency trading strategies, and analyzing financial and social relations between investors and firms. Typical methods include discovering frequent patterns, sequential patterns, graph patterns, network patterns, tree patterns, knot patterns, interactive patterns, and combined patterns.

Kernel learning methods describe, represent and analyze the distributions and numerical relations between economic-financial indicators by individual or multiple (linear or nonlinear) kernel functions and their couplings. Typical methods include linear and nonlinear kernels, vector (space) kernels, tree kernels, sequence kernels, support vector machines, Fisher kernels, spectrum kernels and multi-kernel learning. Such methods can characterize the inter-dependency between micro-market variables such as security price and volume and between macro-economic indicators such as GDP values, exchange rates, and gold prices by multi-distribution kernel functions.

Event and behavior analysis characterize the occurrences, driving forces, evolution and consequences of events, activities and behaviors undertaken by or on an economic-financial object, and detect and predict abnormal, unexpected and changing events and behaviors. Examples are detecting exceptional co-movement between an underlying equity, derivative products, and the equity company’s announcement releasing activities by coupling learning and coupled behavior analysis of their activity sequences. Typical methods including historical event analysis, sequence analysis, Markov chain process, nonoccurring behavior analysis, high-impact behavior analysis, high-utility behavior analysis, and group behavior analysis.

Document analysis and NLP

recognize, identify, extract, summarize, and classify sentiments, opinions, financial events, and entities. They can support topic-oriented modeling, sentiment-oriented analysis, searching, counting and comparison of rule-violated, suspicious, malicious, fragile, deceptive and risky statements, claims, announcements, opinions, sentiments or entities within and between economic-financial documents, reports, and news. Examples are identifying and comparing problematic descriptions, revenue and budget statistics, misleading or misclassified reporting and review comments across monthly financial review reports by Transformer and BERT-derived neural models. Typical methods include language models, named entity analysis, case-based reasoning, sequence labeling, statistical language models, context-free parsing, logical and dependency semantics, distributional and distributed semantics, topic modeling such as latent Dirichlet allocation, word embedding, Bayesian networks, and neural models such as Transformer and BERT models.

Model-based methods

characterize, represent and analyze economic and financial problems in terms of given hypotheses and models. Examples are modeling trading behaviors, market movements, institutional trading, price and index trend, and the influence of trader’s sentiment on the dynamics of financial indicators. Typical models include numerical computation models such as kernel functions, statistical models such as probabilistic graphical models and Bayesian networks, expectation-maximization models, the models for clustering, classification and semi-supervised learning, deep reinforcement learning (DRL) models, and deep neural models.

Social media analysis characterizes, detects, classifies, clusters and predicts linkages, interactions, feedback, sentiment and networking behaviors between entities of economic-financial systems. Such methods model economic-financial systems as social networks by involving networking data from relevant social media, mobile apps, and instant messaging platforms to characterize, quantify and trace the sentiment and opinion and their influence on a security price movement or a market index movement by short text analysis and network analysis. Typical methods include topic modeling, sentiment analysis, emotional analysis, influence analysis, and modeling linkage and interactions.

Iv-D Computational Intelligence Methods

Computational intelligence methods model the working mechanisms of economic-financial systems, investment analysis, economic forecasting, portfolio analysis, and inflation prediction, etc. Such methods are inspired by natural, biological and evolutionary systems, and categorized into the following three classes.

Neural computing methods model the relations, structures, sequential movements, transitions and influence between economic-financial variables by neural networks (NN), in particular, deep neural networks (DNN) nowadays. Examples are modeling the coupled dependencies and dynamics between a stock price movement and macroeconomic variables such as petrol and gold prices by recurrent neural networks (RNN). Typical methods include various artificial neural networks, wavelet neural networks, genetic neural networks, fuzzy neural networks, and DNN.

Evolutionary computing methods

characterize, simulate, analyze and optimize the working mechanisms, evolution, interactions, variances, performance and risk of economic-financial systems inspired by biological and evolutionary models. Examples are simulating and optimizing the development and evolution of a newly released financial product in a market by a particle swarm organization (PSO) model, and estimating and optimizing security price, market index, exchange rate and inflation rate, and credit profiling. Typical methods include ant algorithms, self-organizing map, genetic computing/programming, artificial immune systems, gene expression programming, particle swarm optimization, swarm intelligence, and neural-genetic algorithms.

Fuzzy set methods characterize the relationships, distributions and structures of economic-financial systems in terms of fuzzy set and fuzzy system theories. Examples are modeling market momentum, capital cost, risk, financial solvency, financial structures, and relations between costs on profit and between financial structures and capital costs. Typical methods include fuzzy set theories, fuzzy systems, fuzzy logic, fuzzy neural networks, and genetic fuzzy logic models.

Iv-E Modern AIDS Methods

Here, we discuss several categories of modern AIDS methods, which have been focused for the economic-financial problem-solving in the recent decade or so. They are representation learning, short and informal text analysis, optimization, reinforcement learning, and deep learning methods.

Representation learning methods describe, characterize and model the intrinsic processes, interactions, relations, structures, distributions and characteristics of economic-financial systems, products or problems. Examples include establishing a probabilistic, mathematical, graph-based, network-based, tree-like or neural representation of the assets, participants, and role interactions in a derivative market, and representing stocks by involving tick-by-tick data and external news and social media data about the stocks. Typical methods include probabilistic models, graph networks, network embedding, tree models, neural embedding models, and adversarial learning methods.

Short and informal text analysis

collects, extracts, recognizes, analyzes and classifies short text in social media, SMS and instant messaging systems about economic-financial institutions, products, services, trends, news or participants. Such information is alternative to core data available in markets and economic-financial institutions, but consists of important messages. It can be used for various purposes, e.g., predicting the trend and movements of price, index and exchange rate, extracting and representing keywords, phrases and expressions in a social and virtual community about manipulating a commodity price, and predicting the market sentiment, reputation and confidence on a product or service. Typical methods include explicit short text understanding such as conceptualization for segmentation and labeling, term/tag/phrase similarity learning, dependency parsing, syntax structure analysis, short text classification, query and recommendation; and implicit short-text analysis, e.g., word/phrase/sentence embedding, contextual embedding, short-text conversation (e.g., chatbot), question/answering by neural models such as long short-term memory, attentive RNN, neural models with multi-head attention, encoder-decoder, and variational autoencoder, Transformer, etc.

Optimization methods

model economic-financial problems as optimization problems or apply optimization methods to characterize and analyze optimal economic-financial solutions. Examples include optimizing portfolio design and strategies and return on investment w.r.t. relevant micro- and macro-level financial indicators, optimizing the VaR of a cross-market portfolio with transaction cost, and optimizing algorithmic trading strategies. Typical methods include linear programming, nonlinear programming, quadratic programming, stochastic programming, dynamic programming, multi-objective evolutionary computing, swarm intelligence, information theory, and Bayesian optimization.

Reinforcement learning methods model an economic-financial system as a reinforcement learning problem by optimizing actions, policies and rewards. Examples include modeling the portfolio management of derivative products to optimize the trading objectives, forecasting the price and trend of foreign currencies, modeling property market supply and demand, optimizing order routing in foreign exchange and features markets, and optimizing trading agent actions, trading behaviors, option valuation, and trading and asset management. Typical methods include classic reinforcement learning such as on policy learning, e.g., Bellman Equation, on-policy actor-critic reinforcement learning and Markov dynamic process reinforcement learning, and off policy learning e.g., Q-learning and deterministic policy gradient. Recent focuses have been on deep reforcement learning (DRL) for EcoFin problem modeling, such as deep Q-network (DQN), double DQN, -greedy policy Q-learning, -greedy policy double Q-learning and weighted Q-learning, deep deterministic policy gradient, hierarchical reinforcement learning, recurrent reinforcement learning by integrating RNN with Q-learning, and adversarial learning in reinforcement learning.

Deep learning methods

represent financial variables and their hidden relations and structures in economic-financial systems by multi-layer abstractions and hidden representations. Deep learning can be used for stock representation, asset representation, portfolio representation; modeling sequential trading behaviors and strategies; modeling sequential trading; predicting risk, profit and return of a portfolio; analyzing the couplings, interactions and influence between macroeconomic and microeconomic variables; detecting abnormal movements and financial activities in markets and banking; discovering incompliant behaviors and events in financial reports; understanding customer sentiment and feedback for marketing; supporting smart crowdfunding and crowdsourcing; modeling the influence of external events and news on market movement; and analyzing and predicting the likelihood of financial crisis, vulnerability, and cross-market influence. Typical deep models including convolutional neural networks (CNN), RNN, attention networks, memory networks, generative adversarial networks (GAN), encoder-decoder, autoencoder, graph neural networks (GNN), deep Bayesian networks and neural language models (e.g., Transformer) can be used for the above economic-financial purposes.

Iv-F Hybrid AIDS Methods

Hybrid methods and ensemble methods model economic-financial systems and problems by multiple complementary AIDS techniques through their hybridization, ensembling or synthesis. Such approaches are typically used to address (a) single yet complex economic-financial problems that cannot be well handled by single models; and (b) mixed problems that require multiple methods. Accordingly, there are two main directions of synergy. On one hand, the aforementioned AIDS techniques may be selectively integrated in parallel, sequential or hierarchical ways to enhance their capacity and ability.

Parallel hybridization, also called ensemble, integrates multiple same or different models to address the underlying problems. Examples include combining evolutionary computing and neural computing to generate evolutionary neural models to simulate and analyze a capital market; integrating complexity science and game theory to model and simulate blockchain and cryptocurrencies; training a deep Bayesian network composed of Bayesian dependency learning and hierarchical networks; and learning a copula GNN handling both dependency and connectivity.

Sequential and hierarchical hybridization

integrates multiple methods sequentially or hierarchically to address underlying problems. Examples are to model stock price relations by time series analysis and then predict the abnormal movement by a decision tree classifier; model the micro/macro-economic variable dependency by a copula method and then forecast the VaR by a time series model; hierarchically represent market variables as input and then model the trading behaviors in a market by a RNN before the prediction of market movement. Parallel, sequential and hierarchical hybridization have been widely explored in the relevant communities.

Cross-disciplinary hybridization integrates economic theories and financial theories with AIDS methods, which has been booming in recent years. Examples include migrating multivariate time series to deep sequential models, combining copula methods with deep neural models, integrating autoregression with DNN, learning high-utility behavior sequences, training actionable trading agents and strategies for learning trading history and satisfying portfolio optimization objectives (e.g., VaR). This direction presents a more genuine yet challenging approach for addressing significant economic-financial complexities.

Behavioral economics and finance is a typical area that integrates multiple disciplines of thinking, theories and methods (e.g., psychology, neuroscience, applied behavior analysis, and computational intelligence) to understand, interpret and model the less rational to irrational investment, mispricing behaviors, market inefficiency and market hiddenness that drive market decisions and evolution. It migrates from the hypothetical rationality, bounded rationality and efficiency and transparency in classic economics and finance (i.e., System 1) to the less rational and irrational human choices, market inefficiency (System 2) in real-life human decision-making. Typical behavioral economics and finance methods include prospect theory, nudge theory, natural experiments, and experimental economics. More advanced directions are to incorporate AIDS-driven proactive, personalized, sequential, contextual, online and dynamic modeling, analytics, simulation, learning and optimization into behavioral economics and finance to understand, analyze and interpret human cognitive emotion, mental activities and intention, thinking, decision-making and action-taking in markets and financial businesses.

V Conclusions

This review briefs a comprehensive picture and promising potential of AI-enabled smart FinTech. The new generation of AI, in particular data science, machine learning and deep learning, drives the era of data and intelligence-driven economy and finance, where tremendous opportunities emerge in transforming economic-financial theories, research and practice and advancing AI for tackling real-life significant economic-financial challenges and complexities and delivering actionable intelligence-driven economy and finance.

Vi Acknowledgments

This work is supported in part by Australian Research Council Discovery Grant (DP190101079) and ARC Future Fellow grant (FT190100734).


  • [1] H. Arslanian and F. Fischer, The Future of Finance: The Impact of FinTech, AI, and Crypto on Financial Services.   Palgrave Macmillan, 2019.
  • [2] L. Cao, “Data science: challenges and directions,” Commun. ACM, vol. 60, no. 8, pp. 59–68, 2017.
  • [3] Y. Hilpisch, Artificial Intelligence in Finance.   O′Reilly, 2020.
  • [4] Y. Qi and J. Xiao, “Fintech: AI powers financial services to improve people’s lives,” Commun. ACM, vol. 61, no. 11, pp. 65–69, 2018.
  • [5] L. Ryll, M. E. Barton, B. Z. Zhang, R. J. McWaters, E. Schizas, R. Hao, K. Bear, M. Preziuso, E. Seger, R. Wardrop, P. R. Rau, P. Debata, P. Rowan, N. Adams, M. Gray, and N. Yerolemou, “Transforming paradigms: A global ai in financial services survey,” 2020.
  • [6] V. Dhar and R. M. Stein, “Fintech platforms and strategy,” Commun. ACM, vol. 60, no. 10, pp. 32–35, 2017.
  • [7] B. Nicoletti, Future of FinTech.   Palgrave Macmillan, 2017.
  • [8] L. Cao, G. Yuan, T. Leung, and W. Zhang, “Special issue on ai and fintech: The challenge ahead,” IEEE Intelligent Systems, vol. 35, no. 2, pp. 3–6, 2020.
  • [9] C. L. Dunis, P. W. Middleton, A. Karathanasopolous, and K. Theofilatos, Artificial Intelligence in Financial Markets.   Springer, 2019.
  • [10] International Telecommunication Union, “Assessing the economic impact of artificial intelligence,” 2018, issue Paper No. 1, September 2018.
  • [11] B. Li and S. C. H. Hoi, “Online portfolio selection: A survey,” ACM Comput. Surv., vol. 46, no. 3, pp. 1–36, 2014.
  • [12] T. Lynn, J. G. Mooney, P. Rosati, and M. Cummins, Disrupting finance: Fintech and strategy in the 21st century.   Palgrave Pivot, 2019.
  • [13] B. G. Buchanan, “Artificial intelligence in finance,” Tech. Rep., 2019.
  • [14] G. Cornuéjols, J. Peña, and R. Tütüncü, Optimization Methods in Finance (2nd ed.).   Cambridge University Press, 2018.
  • [15] N. Ehrentreich, Agent-Based Modeling.   Springer, 2008.
  • [16] J. E. Gentle, W. K. Hardle, and Y. Mori, Handbook of Computational Finance.   Springer, 2012.
  • [17] M. Gilli, D. Maringer, and E. Schumann, Numerical Methods and Optimization in Finance.   Academic Press, 2019.
  • [18] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning.   The MIT Press, 2016.
  • [19] L. Hamill and N. Gilbert, Agent-Based Modelling in Economics.   John Wiley & Sons, 2016.
  • [20] J. B. Heaton, N. G. Polson, and J. H. Witte, “Deep learning for finance: Deep portfolios,” Applied Stochastic Models in Business and Industry, vol. 33, pp. 3–12, 2017.
  • [21] C. Kearney and S. Liu, “Textual sentiment in finance: A survey of methods and models,” International Review of Financial Analysis, vol. 33, pp. 171–185, 2013.
  • [22] B. Kovalerchuk and E. E. Vityaev, Data Mining in Finance: Advances in relational and hybrid methods.   Kluwer Academic Publishers, 2000.
  • [23] T. L. Meng and M. Khushi, “Reinforcement learning in financial markets,” Data, vol. 4, no. 3, pp. 110:1–17, 2019.
  • [24] G. Mitra and L. Mitra, The Handbook of News Analytics in Finance.   Wiley, 2012.
  • [25] A. M. Özbayoglu, M. U. Gudelek, and O. B. Sezer, “Deep learning for financial applications : A survey,” CoRR, vol. abs/2002.05786, 2020.