What Do Banks Evaluate When They Screen Borrowers? Soft Information, Hard Information and Collateral

See, for example, T. Sugiyama, “Loan portfolio management: Reflect credit risk in loan interest rates and maximize profit opportunity,” Nikkei Financial Daily, May 22, 1996 (in Japanese).

The original articulation of the collateral principle was in the context of corporate bond issuance prior to World War II. The principle was described as a response to the Showa Depression in Japan during the period 1930–1931 (Hoshi and Kashyap 2001, p. 30).

See Harmann (1976) for a discussion of factor analysis. Factor analysis assumes a linear relationship between observed variables and a small number of unobserved variables (called common factors), which are considered to drive the observed variables. It then identifies the common factors to reproduce the correlations among the observed variables as much as possible. The best-known application of factor analysis in the field of finance is Arbitrage Pricing Theory of Ross (1976).

A similar criticism applies to other studies including Berger et al. (2005), García-Appendini (2007), and Berger and Black (2010), in the sense that they principally rely on observed contract terms and/or the strength of bank-firm relationships when they identify lending technologies as opposed to the factors that drive these contract terms.

The questionnaire for this survey (in Japanese) is available in Yamori (2006, p. 96–110).

The main banks are identified by the responding firms themselves.

The Credit Guarantee Corporation is a government financial institution that guarantees commercial loans.

Judging from the original questionnaire wording, it is not clear to what extent (or fraction) Criterion 17 reflects hard and/or soft information (see the precise translation in Table 2). However, because it is collected by only some institution(s) (e.g., credit reporting agencies), the information is highly likely to be collected in an organized manner. In addition, by definition it is difficult to document and transfer soft information (especially for large institutions). It is therefore highly likely that Criterion 17 measures hard information such as Dunn and Bradstreet-type trade credit performance indices.

Criterion 22 is distinct from the other criteria because it does not directly relate to the characteristics of borrowers. As it pertains to the credibility of auditors with the banks, however, we can infer that the criterion indirectly represents soft information on borrowers.

Yosano and Nakaoka (2010) use survey-based information about the checklists that banks use when making loan decisions. However, this is bank-level, not borrower-level, information, and their focus is on bank performance.

For example, Berger and Black (2010) identify several types of fixed asset lending based on information about whether the loan is secured or not.

The latter is the approach adopted in Berger and Black (2010) and García-Appendini (2007).

For factor analysis, see the standard and definitive textbook of Harmann (1976).

The first five of the 22 eigenvalues are 5.28, 3.11, 1.86, 1.23 and 1.13

Although it is difficult to discern to what extent these criteria reflect hard vs. soft information about borrowers, we can at least infer that Criteria 16 and 17 represent hard information, while Criteria 21 and 22 represent soft information. See also Footnotes 8 and 9 as well.

Note that as explained above, the original variables for these criteria (i.e. the observed categorical scores) measure the extent to which (or how much) the main bank emphasizes these financial ratios when underwriting loans. In other words, a low score means that the bank ignores the ratio as being important in underwriting (regardless of whether it is a strong or weak ratio).

A large bank (LARGE) in our sample is either a city bank, a long-term credit bank or a trust bank. In addition to being the largest banks, these tend to provide a broad range of services, including investment and international banking, and to have complex organizational structures. A bank operating regionally (REGIONAL) is either a first- or second-tier regional bank. These banks are typically medium-sized and do not operate nationwide. Their business areas are usually located in or around one of the 47 prefectures in Japan where their main branch office is located. Finally, a UNION bank is either a Shinkin bank or a credit cooperative. Both forms are cooperatives, usually small in size and not-for-profit. See Uchida and Udell (2010) for a discussion of the types of banks in Japan.

Berger et al. (2005) and Uchida et al. (2008a) find supporting evidence for the Stein hypothesis based on a methodology that is different from ours.

See Degryse et al. (2009) for an extensive review of the empirical literature.

As a robustness check, we also run the regressions using the scores calculated from the factor analysis on the “very much” dummies (see Section 3.1). The qualitative results remain unchanged, although fewer variables are significant. This is possibly because of a loss of information from not differentiating answers other than for “very much.”

A positive coefficient of log(ASSET) is also inconsistent with the relationship lending hypothesis, but its impact is economically insignificant because a 10% increase in asset increases SCORE_REL by only 0.0027, which is far smaller than the standard deviation of SCORE_REL (= 0.9).