基于神经网络的个人信用模型

【作者】 胡勇；

【导师】 黄纯国；

【作者基本信息】 西南财经大学 ， 金融学， 2006， 硕士

【摘要】 随着我国加入WTO后金融市场的逐步对外开放,国外银行业将大举进入中国,它们具有在国际竞争中培养出来的高超的风险控制技术和能力,可以提供丰富多样化的金融产品以满足市场的需求和采用先进的管理制度。而这些恰好是我国银行业的薄弱环节,国内的银行业在未来将面临国外银行业越来越强大的竞争压力。与此同时,国内的金融市场正在逐渐的发育成熟,资本市场正在日益的完善,越来越多的国内企业通过国内外的资本市场而不是银行借贷来筹集资金,这就意味着国内银行业的主要业务和主要的盈利来源——公司业务,正在逐渐失去最好和最优良的客户,为了争夺客户银行不得不提供比以前更优惠的贷款条件和将目标转向风险更大的次优客户。银行将面临着越来越大的风险,但利润却越来越低。因而为了控制风险和寻找新的利润来源,银行不得不拓展新的业务,同时监管当局对资本的要求也日趋严格。个人消费信贷业务在此背景下成为了国内银行业关注的焦点。金融统计数据表明, 2005年1-4月,国内个人储蓄占金融机构各项存款的比例高达51.3%,储蓄存款成为金融机构存款来源的半壁江山。2005年4月末,金融机构个人贷款余额达到27043亿元,比上年同期增长了57%,呈现了快速发展的态势。并且随着我国社会经济的不断发展,中产及富裕阶层不断扩大,这个为发展个人消费信贷业务奠定了物质基础。如汽车消费和住房消费市场的启动和繁荣就为个人消费信贷业务中的消费贷款开辟了广阔的空间。个人消费信贷业务具有低资本消耗、低风险、收益稳定的特征,其中的信用卡、住房信贷等业务,不仅能使银行获得稳定的利差收入,还可以获得可观的中间业务收入,这将弥补因利差或市场萎缩而带来的利润下降因素的影响,而且由于个人消费信贷业务风险权重低,能够有效减轻资本充足率带来的压力。随着国内银行业对个人消费信贷业务的日益重视,个人消费更多还原

【Abstract】 Nowadays the personal credit operation becomes the focus of China’s bank industry. The risk in the personal credit risk is the most serious risk for the commercial banks. Thus the analysis of the personal credit risk is the centre of the control of it, which is of great significance for the operation safety of the commercial banks. Aiming at the weakness of the personal credit analysis in China, the thesis generally researches the personal credit theory and analysis and compares the current international researches. Based on which the thesis combines the mathematic approach and model-designing theory. And works out the grading model for personal credit with Neural Network.The thesis consists of five chapters.Chapter one is about the basis of the personal credit researches, such as the definition, concept, characteristics, and functions of personal credit. Based on which the definition of the personal credit risk is given.Chapter two is the introduction of the meanings and approaches of the personal credit evaluating.The advantages and advantages are of different approaches compared.Chapter three is the basic theory of the artificial neural network, the origination, development, feature model and related algorithm of which are detailed. Then the thesis compares the artificial neural network with biologic neural network.Chapter four details the basis of the modeling of the thesis-- Error Back Propagation Algorithm, or BP, which is the stress of the thesis. The basic principle of BP is introduced and the studying process of BP is deducted detailedly and its concrete steps are concluded. Meanwhile the improved algorithm to solve the problems of BP as slow constringency and the trap of regional minimum.Finally The thesis detailedly describes the design of neural network of BP. The更多还原