«February 20, 2003 1 An earlier version of this paper entitled ”Herd Behavior between Diﬀerent Types of Japanese Banks” was presented at the ...»
Herd Behavior in the Japanese Loan Market:
Evidence from Semi-Macro Data 1
Ryuichi Nakagawa Hirofumi Uchida
Faculty of Economics Faculty of Economics
Hiroshima University of Economics Wakayama University
February 20, 2003
1 An earlier version of this paper entitled ”Herd Behavior between Diﬀerent
Types of Japanese Banks” was presented at the annual meeting of the Japanese Economic Association (October 13, 2002, Hiroshima University). The authors would like to thank Yosuke Takeda (the discussant), participants in the Monetary Economics Workshop, Ken-ya Fujiwara, Takashi Hatakeda and Keiichi Hori for their helpful comments. Uchida acknowledges the ﬁnancial support from Wakayama University Research Fund.
2 Corresponding author: Faculty of Economics, Hiroshima University of Economics, 5-37-1 Gion, Hiroshima 731-0192, Japan. Phone&Fax: (81)-82-871-1485.
Herd Behavior in the Japanese Loan Market:
Evidence from Semi-Macro Data Abstract The aim of this paper is to examine whether herd behavior exists or not between diﬀerent types of Japanese banks. Using data of loans outstanding by types of banks from 1980 to 2000, we investigate statistical causality between loans from diﬀerent types of banks in Japan. Time-series analysis manifested the causality from city banks to regional banks, from long-term credit banks to city banks, and from trust banks to city banks. Consistent with historical events in Japan, these relationships have been found especially for loans to emerging industries and for those in the 1980s. We also conducted analysis to clarify possible causes of the detected herding. Among diﬀerent stories behind herd behavior, the results are most consistent with that based on inference from other type of banks. The results between long-term credit banks and city banks, and those between trust banks and city banks are also consistent with the ”Cowbell-like” eﬀect.
Keywords: herd behavior, banks, loan market, Japan, Cowbell eﬀect JEL classiﬁcation number: G21 1 Introduction In Japan, the existence of herd behavior among banks has long been pointed out. Until early 1980s, capital markets were underdeveloped. Loans and bond issues underwritten by banks were the only sources of ﬁnance for a great number of ﬁrms. A large amount of funds ﬂow in Japan had therefore been intermediated by banks.1 Under such circumstance, competition among banks was less severe. It was often claimed that loans, fees, salaries, deposit rates, and so on had been set uniformly.
The situation has changed since the disintermediation prevailed. Due to the development of capital markets after the huge issue of government bonds in late 1970s, a wide variety of ﬁnancing sources have become available for large companies in the 1980s and banks in Japan have been facing diﬃculty in ﬁnding alternative sources of income. Nevertheless, we can still ﬁnd some casual evidence which might imply the existence of herding. For example, a conventional explanation for a fair amount of bad loans in Japanese banks is that in the bubble era they did not fully monitor their borrowers and herded to lend, for example, real estate industries. Herding between large and small banks which have close capital relationships is often claimed as well. In ﬁnancial keiretsu, several banks granted loans under the coordination by main banks. In spite of these casual claims, however, there are almost no study that formally investigates whether herd behavior has been observed or not in the Japanese banking industry.
This paper is the ﬁrst fact-ﬁnding study which attempts to examine whether herd behavior exists or not in the Japanese loan market. We use data on loans outstanding by types of banks from 1980 to 2000 to investigate herd behavior between diﬀerent types of Japanese banks.
The analysis is based upon time-series technique and consists of two stages. First, we will investigate whether any statistical causality exists or not between loans from diﬀerent types of banks, and in which direction, if any. If some causality is detected, we then proceed to the detection of the possible cause of herd behavior. Among possible stories behind herd behavior, we will focus on three stories which can be detected by semi-macro data analysis: (a) accidental herding resulting from the same action based on common information (herding based on common information), (b) intentional herding in which uninformed banks follow informed ones (inference See Hoshi and Patrick for more details.
from informed agents), and (c) herding based on causes other than (a) or (b) (other explanations).2 Introducing macroeconomic variables as explanatory variables and checking if the detected causality is still signiﬁcant, we determine which of the three stories is most likely to be relevant.
The analysis manifests the causality from city banks to regional banks, from long-term credit banks to city banks, and from trust banks to city banks. Consistent with the historical events in Japan, these relationships are most likely to be found for loans to emerging industries and for those in the 1980s. Furthermore, the results are most consistent with type (b) herding, herding based on inference from other type of banks. The results between long-term credit banks and city banks and those between trust banks and city banks are also consistent with the ”Cowbell-like” eﬀect of loans from long-term credit or trust banks.3 The analysis in the present study is similar to that in Jain and Gupta.4 Their study takes a similar method of approach using semi-macro time-series data. However, the focus of their study is diﬀerent from ours. They examine herd behavior among U.S. banks with respect to the loans to Latin American countries from late 1970s to early 1980s. Furthermore, they focus only on whether any herd behavior would have existed or not, while we further try to identify which type of herd behavior would be the most relevant. Hence, the present study has technical advantage over theirs.5 So far, at least to the best of our knowledge, the only study which investigates herd behavior in the Japanese banking industry is Miyamoto[20, ch.8].
However, his focus is on the deposit market, and the analytical method is completely diﬀerent from ours.6 On possible causes of herding, see Section 2.
The original Cowbell eﬀect is the eﬀect that loans from the Japan Development Bank induced other banks’ loans to the same borrower. This eﬀect was ﬁrst claimed by Higano. See Section 3.3.1 for more detail.
There are other studies which examine herd behavior in diﬀerent situations. Examples are; herding among fund managers (Lakonishok et al., Grinblatt, Titman, and Wermers, Wermers, inter alia), among security analysts (Graham, Hong, Kubik, and Solomon, inter alia), and among macroeconomic forecasters (Lamont, Laster, Bennett, and Geoum, inter alia). These studies use detailed individual data on managers, analysts, and forecasters, respectively. In contrast to these studies, the present study aims at detecting herding at a semi-macro level.
In the terminology of the present paper, Jain and Gupta estimated only the equations without macroeconomic variables.
Following a method similar to Lamont, Ashiya and Doi ﬁnd herd behavior by The remainder of this paper is composed as follows. Section 2 summarizes the reason herd behavior can take place. In Section 3, we introduce the data and the method of approach employed in this paper. Section 4 reports the results. Finally, Section 5 concludes the paper.
2 Possible causes of herd behavior In theoretical studies, a number of explanations have been given on herd behavior.7 Among others, the present paper will focus on the following three explanations, since these can be identiﬁed, at least to some extent, using semi-macro data.8 (a) Herd behavior based on common information According to this explanation, herd behavior takes place not by intention but by accident. Reacting to the same information, diﬀerent agents can accidentally act in the same way. For example, prospective information about a certain industry may give rise to simultaneous increases in diﬀerent banks’ loans to that industry. Such information is not restricted to public one. Common private information may cause herd behavior. As indicated by Lakonishok, Shleifer, and Vishny, this type of herding may even be eﬃcient.
Japanese macroeconomic forecasters.
See Bikhchandani, Hirshleifer, and Welch, and Devenow and Welch for survey.
Other explanations can be given on herd behavior. First, Falkenstein shows that herd behavior is observed among those mutual funds that have similar comparative advantage. By analogy of this reasoning, herd behavior can be found among banks with similar characteristics, i.e. banks of the same type. For example, long-term credit banks in Japan had had comparative advantage in long-term ﬁnance so that they might have shared a tendency to lend intensively to, for example, heavy industries. However, since the data to be used in this paper is aggregated by each type of banks, it is impossible to detect this type of herding in the present study.
Second, some kind of payoﬀ externality may account for the reason herding takes place.
Examples are: bank run (Diamond and Dybvig inter alia), liquidity (Devenow and Welch), and information production (Froot, Scharfstein, and Stein and Hirshleifer, Subrahmanyam, and Titman). However, it is not probable that this type of herding takes place between diﬀerent types of banks.
Finally, in the presence of career concern by managers, herding may take place so as to obtain higher reputation (Scharfstein and Stein). This type of herding should not be found in semi-macro data used in the present study, either.
(b) Inference from informed agents Agents with inferior information may follow those with superior ones. This type of herding is also likely to be observed in Japan. Banks in Japan had been segmented in their activities by the regulation on the segregation of business areas in ﬁnancial industries.
Small and large banks had had an advantage in lending to small and large ﬁrms, respectively, while long-term credit and trust banks in long-term ﬁnance. After the relaxation of this regulation, business area of each type of banks began to overlap. It is therefore possibly the case that banks of a certain type had started lending to a new class of customers by following banks of other type who already had advantage in lending that class.
(c) Other explanations If neither type (a) nor type (b) herding is detected, several explanations are possible. Since we will use semi-macro data, these explanations cannot be identiﬁed from each other. Thus, we collectively call them (c) other explanations. Included among them, for example, is socalled the cascade (Bikhchandani, Hirshleifer, and Welch and Banerjee).
Consider a situation in which multiple agents have to make a similar decision sequentially based on private information as well as past records of decisions made by their predecessors. The authors show that agents of later turns tend to be more ignorant of their private information and more frequently mimic their direct predecessors.9 Also included is herding based on irrational behavior. Although explanations thus far a re all based upon rational behavior, herding may be brought about simply by irrational behavior. This is indeed one of the most commonly indicated causes of herding in the ﬁnance literature.10 3 Data and Methodology 3.1 Data We use data from Financial and Economic Statistics Monthly by the Bank of Japan, Financial Statements Statistics of Corporations by Industry (Quarterly) by the Ministry of Finance, Monthly Statistics Report by Tokyo Stock Exchange, some of which are obtained from Thomson Financial Datastream For more on the cascade, see Bikhchandani, Hirshleifer, and Welch and Gale.
This type of herding is also called ”herding on noise”. See, for example, Shiller.
Database. The main data is quarterly ”Loans and Discounts Outstanding by Industry” in Financial and Economic Statistics Monthly, by which we can identify the amount of loans granted to diﬀerent industries by diﬀerent types of banks.11 The sample period is from 1980 to 2000. This includes the period of ﬁnancial liberalization and the bubble period in Japan. Since some structural change may have taken place, we will focus mainly on two sub-sample periods, the 1980s and the 1990s.1213 Included in the types of banks are (i) city banks, (ii) regional banks, (iii) second regional banks, (iv) long-term credit banks, and (v) trust banks.14 City banks are large and operate nationwide. Regional banks and second regional banks are typically small- or medium-sized banks which usually focus on regional ﬁnance. Long-term credit banks and trust banks are (were) the banks that have (had) special aims: The former was for long-term ﬁnance and the latter for trust services. Both types of banks have often been characterized as the main providers of long-term funds in Japan. Note that this structure of banking industry in Japan has been changing drastically these days, especially following the ﬁnancial system reform so-called the big-bang in the mid-1990s.15 The industries included are (1) manufacturing, (2) construction, (3) electricity, gas, heat supply and water (hereafter electricity), (4) transport and communication (hereafter transport), (5) wholesale, retail trade, eating and drinking places (hereafter wholesale), (6) ﬁnance and insurance, (7) real esOne might wonder why we do not use ”New Loans for Equipment Funds by Industry” (ﬂow data). This data is available only after 1993 and only for broader industry classiﬁcation. Furthermore even with stock data, we can detect herding in the adjustment of the stock of loans, which will be shown below.
Although some data are available before 1980, they are insuﬃcient to conduct analysis with regard to the whole 1970s. Furthermore, due to the reason to be explained in Section 3.3, it is appropriate to focus on the period from 1980 on, after which the window guidance by the Bank of Japan was alleviated.