1、1 中文 3620 字 本科毕业论文（设计） 外 文 翻 译 外文出 处 University Robert Schuman, Strasbourg III InstitutdEtudes Politiques 外文 作者 Brigitte Godbillon 原文 ： Credit Risk Management in Banks: Hard Information, Soft Information and Manipulation Introduction Information remains a crucial input for the banking industry. Bank
2、s are confronted to informations asymmetry problems because of borrowers informational opacity. This opacity varies borrowers type, SMEs being considered as the most opaque (because of lack of public information). In order to resolve this informational asymmetry, the bank can acquire two types of in
3、formation: hard information, which is external, via public information (balance sheet data, rating, scoring . ), and soft information, which is internal, via bank-borrower relationship (judgement, opinions, notes, reports . . . ). This also implies two lending technologies : transaction lending vers
4、us relationship banking. A recent stream of literature puts forward distinctions to be made between hard and soft information (Petersen, 2004)1. Taking into account soft information in risk analysis can increase estimations precision of borrowers quality (Lehmann, 2003； Grunert et al., 2005), but ha
5、s the disadvantage of being non verifiable and therefore manipulable. This type of information can influence credit risk management in banks, but may also have an impact on banks organizational structure, which should be adapted to soft information in order to avoid the 2 consequences and costs of i
6、ts manipulation. Recent research on risks management in banks puts forward the importance of informations treatment. Hakenes (2004) considers the banker as a “specialist” of informations treatment and risks monitoring. Danielsson et al. (2002) analyze banks choice of risks management system, investi
7、gating different levels of power delegation implying more or less transmission of information. However, this stream of research doesnt distinguish hard and soft information. Therefore, this article investigates the impact of informations type on the balance sheet structure and organization in terms
8、of credit risk management of the bank. We propose a theoretical model of the credit decision within a principal-agent framework with a bank director (or a director of the credit risk department) and a credit officer (or a banks agency clerk). The director allocates equity for Value at Risk coverage.
9、 He also decides on the officers budget, as well as her wage, which are both a function of a signal, based on hard information only or a combination of hard and soft information. The difference between the two types of signal lies in their nature, more precisely their verifiability and manipulabilit
10、y, as well as their level of precision. A combination of hard and soft information is more precise than hard information only, but is not verifiable by the director, as the soft component is manipulable. Soft information is therefore a source of moral hazard with hidden information. It is a potentia
11、l driver of organizational modifications in the bank in order to limit te moral hazard problem. We find several interesting results. We show that taking into account soft information in risk management can allow to economize equity for VaR coverage, under certain conditions. However, we also verify
12、the existence of the soft informations manipulation incentive by the officer. We then propose a wage scheme to impeach this manipulation. The influence of soft information on banks organizational structure is modelled through a specific salary package. The comparison of solutions from the two framew
13、orks (one with hard information only versus a combination of hard and soft information), realized with numerical simulations, confirms that the soft informations component can provide an advantage, 3 as it effectively allows to reduce equity for VaR coverage. The rest of this article is organized as
14、 follows. In section 2, we present elements allowing to distinguish between hard and soft information, as well as recent theoretical research investigating informations type influence on banks organizational structure. The credit risk decision model is presented in section 3. Sections 4 and 5 provid
15、es results in the hard informations case, and deduce pros and cons of soft information, in particular the existence of a manipulation incentive in presence of a soft information based signal. An incentive wage scheme for the officer is then proposed to resolve the manipulation problem in section 6.
16、Finally, the results from the hard and a combination of hard and soft information cases are compared, using numerical solutions, and presented in section 7. Section 8 concludes the article. Regarding soft information, its most particular characteristic is to be tightly linked to the environment and
17、context where it was produced. In the banking framework, this environment is the bank-borrower relationship, which, through multiple interactions in time, gives access to private and confidential information, superior to publicly available information (Berger, 1999； Boot, 2000； Berger and Udell, 200
18、2； Elsas, 2005). Thus, soft information has the advantage to increase the predictive capacity of hard information, but remains non verifiable. The latter makes this type of information easily manipulable by the agent responsible of its production and treatment, and therefore imposes a particular org
19、anizational structure. Soft informations capability to increase hard informations predictive power is documented by empirical research which aims at investigating qualitative factors impact on default risk prediction. These studies use in particular banks internal ratings which are integrated into d
20、efault risk prediction models, along with hard balance sheet and financial factors. A strong part of these internal rating are based on qualitative factors, and therefore soft information, as management quality or business perspectives of the borrower 4. The integration of qualitative factors into default risk prediction models increase their discrimination and reclassification performance, and therefore default prediction accuracy( Lehmann, 2003； Grunert et al., 2005). Also,qualitative factors appear to be less dispersed and more stable.
