Page 50 - CCFA Journal - Second Issue
P. 50
加中金融
6. Recommendations to accelerate the DRM
随着互联网快速普及,全球数据呈现爆发式增长,每次浏
览网页、搜索或者用智能手机上网,几乎都会增加数十亿 A. Construction of big data platform of risks
字节之多的数据,而且数据结构多样、逻辑复杂,传统的 With the rapid spread of the Internet, so is the global data
关系型数据库已渐渐无法满足风险管理的数据分析需求。 explosively expanded. The data structure is diverse with
在此背景下,建立了由关系型数据库、非关系型数据库、 complex logic. Thus the traditional relationship databases have
Hadoop 集群、Spark 计算框架以及机器学习服务器组成的 gradually been unable to meet the data analysis needs for risk
风险大数据平台。通过高维机器学习模型和集成学习模型, management. So a big data platform for RM consisting of
量化企业的风险,同时,建立网络关系特征,识别企业间 relationship and non-relationship databases, Hadoop clusters,
关联关系,对企业风险舆情信息的精准识别;使用机器学 Spark computing framework, and ML server is established. It
习,结合大数据,不断迭代训练,完成开发新的预警模型 quantifies enterprise risk through high-dimensional ML model
和新的评级系统。不断夯实基础数据,一方面,实施数据 and integrated learning model. At the same time, to establish
整合,扩大数据源,提升数据价值。内部整合信贷、会计、 network relationship characteristics, to identify inter-enterprise
资产负债、审计等系统数据,实现企业内数据的全面调用 correlation relationship, accurate identification of enterprise
和挖掘;外部引入工商、司法、海关、公积金、行业协会、 risk public opinion information, to use machine learning,
互联网舆情等数据信息,实现外部数据汇总;同时大力提 combined with big data, continuous iterative training, to
升数据分析、数据清洗、数据挖掘能力,从海量数据中去 complete the development of new early warning model and new
除无效数据,提高数据质量和价值。另一方面,实施数据 rating system.
融合。打破内外部信息壁垒,最大限度集成数据,构建覆 B. Promote the construction of specialized and intelligent
盖客户、行业、区域、人员的多维度、多时点信用风险数 models.
据视图。
The Internet generates a lot of user data every day, search and
(2) 推进专业化、智能化模型建设 recommendation models need to be optimized continuously and
frequently, self-iteration frequency is faster and more accurate
互联网每天都生成海量用户数据,搜索、推荐模型需要持 than in the financial field, through ML can solve the problem of
slow manual iteration of models. In financial RM, ML models
续频繁地优化,自迭代频次比金融领域更快、更准确,通 can effectively and quickly iterate through monitoring of model
过机器学习可以解决模型人工迭代慢的问题。在金融风险 characteristic performance, lending groups, and business
管理中,通过对模型特征性能、借贷群体和业务反馈等多 feedback. Build risk control models using AI technology and
方面的监控,机器学习模型能有效地快速自迭代。通过使 apply them to subdivided business processes such as credit
用人工智能技术构建风控模型,并将模型应用到如授信定 pricing, pre-credit review, post-credit monitoring, and
价、贷前审核、贷后监控、交易欺诈侦测等细分业务流程 transaction fraud detection.
中。一方面,金融企业应集合不同客户和业务特点,做好
信用决策模型的研发。如在授信决策方面,构建客户融资 C. Data security database through data sharing.
需求与其风险承担能力相适应的授信决策模型。另一方面, With big data industry alliance, relevant trade associations and
构建多维度的预警监控模型。模型应覆盖行业风险、区域 other organizations, one would establish a shared data security
风险、客户风险、产品风险、案件操作风险等,重点关注 prevention database to promote the collection and sharing of
关联结构传导风险、贸易链传导风险、异常交易与资金流 data security information, data repair initiatives. At key aspects
向、征信信息及征信行为等内容。 of the data lifecycle, such as data generation, acquisition,
transmission, storage, use, sharing, destruction, to comb
(3) 建立数据安全防范数据库,加强数据共享 through the technical tools required for data security, and share
technical inventory in a timely manner to ensure that all parties
以大数据产业联盟、相关行业协会等组织为依托,在大数 have the ability to quickly identify similar technologies when
据生产使用过程中的风险监控和管制以及风险预警和化解 discovering security vulnerabilities or potential threats.
方面,建立一个共享的数据安全防范数据库,促进数据安
全防范信息和修复举措的收集和共享,低成本、高质量、 D. Strengthen the application of fintech R&D and expertise.
高频度地生产、使用数据安全防范相关知识。在数据产生、 The integration of fintech and risk management is a systematic
采集、传输、存储、使用、共享、销毁等数据生命周期的 project, which is faced with the constraints and breakthroughs
关键环节,梳理总结数据安全防护所需具备的技术手段和 of concept, foundation, technology and talents. The foundation
工具,并将技术清单及时共享,确保各方在发现安全漏洞 of fintech is information and data. Financial enterprises should
或潜在威胁时具备迅速找到类似技术的能力,以正确应对 conform to the development trend of information technology,
并降低潜在危害。充分利用先进的自动化工具、信息采集 firmly establish "data to create value" as the focus, data
和智能分析等现代化技术手段,探索建设机器学习型智能 management as a foothold. Fintech empowers RM, but also puts
模型风险管理平台,识别数据管理中的薄弱环节并自动发 forward higher demands for RM talents. In the future, we need
出提示,提高风险监测、反馈、防护和应对的效率。 to invest heavily to and actively cultivate complex fintech
talents with data processing and analysis, new technology
(4) 加强金融科技研发应用和人才培养 application, data model research and development and risk
management ability.
金融科技与风险管理融合是一个系统工程,面临着理念、
基础、技术和人才等方面的制约和突破。金融科技的基础
是信息和数据,金融企业要顺应信息科技发展趋势,牢牢
树立“以数据创造价值”为着眼点,以数据管理为立足点。
要培育数据文化,树立数据意识、数据思维,从数据上看
问题,从数据上分析问题、理解问题、解决问题。金融科
技赋能风险管理,同时也对风险管理人才队伍提出了更高
的要求。未来需要大力度的持续投入,积极培养具有数据
处理分析能力、新技术应用能力、数据模型研发能力和风
险管理能力的复合型金融科技人才。
CCFA JOURNAL OF FINANCE DECEMBER 2020
Page 50 第50页