Download hard and soft information.pdf and more Lecture notes Accounting in PDF only on Docsity!
August 2018
Information: Hard and Soft*
José María Liberti Kellogg School of Management Northwestern University and DePaul University and Mitchell A. Petersen Kellogg School of ManagementNorthwestern University and NBER June 2018 Abstract Information is a fundamental component of all financial transactions and markets, but it can arrive in multiple forms. We define what is meant by hard and soft information and describe the relative advantages of each. Hard information is quantitative, easy to store and transmit in impersonal ways, and its information content is independent of its collection. As technology changes the way we collect, process, and communicate information, it changes the structure of markets, design of financial intermediaries, and the incentives to use or misuse information. We survey the literature to understand how these concepts influence the continued evolution of financial markets and institutions. Keywords: soft information, hard information, hardening soft information, boundaries of firm, organizational design, lending, distance, transmission of information, FinTechJEL codes: G20, G21, G
- This is a significantly revised version of the working paper previously circulated as “Information: Hard and Soft”(2004) by Mitchell A. Petersen. The authors thank the editors Efraim Benmelech and Paolo Fulghieri for their patience and guidance in bringing this paper to fruition. We also thank Sumit Agarwal, Alan Berger, Richard Cantor, BruceCarruthers, Beverly Clingan, Barry Cohen, Kent Daniel, Diane Del Guercio, Bob DeYoung, Joey Engelberg, Scott Frame, Andreas Fuster, Jon Garfinkel, Michael Faulkender, Andrew Karolyi, Juhani Linnainmaa, Tamim Majid,David Matsa, Gregor Matvos, Amit Seru, Philip Strahan, and conference participants at the University of Oregon, the Midwest Finance Conference, the University of South Carolina, and SFS Cavalcade for their suggestions and advice.The research assistances of Sang Kim, Austin Magee, and Mark Scovic is greatly appreciated.
I) Introduction. Information is an essential component in all financial transactions and markets. A major purpose of financial markets and institutions is to collect, process, and transmit information. Given the importance of information and its transmission to the study of finance, as technology changes the way information is communicated, it also fundamentally changes financial markets, securities, and institutions, especially financial intermediaries. However, new technologies (i.e., those developed in the past fifty years) are more adept at transmitting and potentially processing information that is easily reduced to numbers. We call this hard information. Information that is difficult to completely summarize in a numeric score, that requires a knowledge of its context to fully understand, and that becomes less useful when separated from the environment in which it was collected is what we call soft information. Building upon the extensive literature on “soft” and “hard” information, we examine the definitions of these terms and their role in understanding financial markets and institutions. The distinction between soft and hard information arose in the finance literature as a way to understand the evolving organization of lenders, although the theoretical ideas reach back much further. Banks have historically been a repository of information about borrowers’ creditworthiness and the kinds of projects available to them. This information was collected over time through frequent and personal contacts between the borrower and the loan officer. Over time the banks built up a more complete picture of the borrower than was available from public records. This private information, most of it soft information, was valuable to the bank. The value arose not only from its ability to inform the bank’s lending decisions but also due to the difficulty of replicating and transmitting the information outside the bank.
III, we describe the main advantages of each type of information using the literature to provide examples and intuition. In Section IV we return to the roots of the soft and hard information literature. We start with a discussion of its foundation in the theoretical banking and organizational design literature, and then we turn to efforts by the empirical literature to measure information type, directly and indirectly (e.g., by using geographic or organizational distance). This leads us to a discussion of the empirical challenge of designing incentives as a function of the type of information an institution uses. In the next section, we examine applications of soft and hard information outside of the banking literature. Specifically, we examine the lessons learned from the financial crisis as seen through the lens of information type. We also discuss the emerging work on FinTech, which in many ways is the newest attempt by markets and firms to replace soft information with hard. This section provides a guide to the future evolution of this literature, as financial innovation and financial crisis are reoccurring themes in finance. Section VI concludes.
II) Defining Soft and Hard Information. An initial challenge of using soft and hard information as useful constructs has been creating precise definitions. As the literature has expanded, the problem has not gotten easier. Thus, we will start with a brief description of the attributes of information that make it soft or hard. This description should be both consistent with much of the literature and also useful in framing research questions. Like many labels in finance (e.g., debt versus equity), there is no clear dichotomy. Rather than two distinct classifications, we should think of a continuum along which information can be classified. Our interest is what characteristics of information, its collection, and its use make it classifiable as hard or soft, and how these characteristics influence the structure of financial markets and institutions.
A) Characteristics of Soft and Hard Information.
- Numbers versus text. Hard information is almost always recorded as numbers. In finance we think of financial statements, the history of payments which were made on time, stock returns, and the quantity of output as being hard information. Soft information is often communicated as text. 1 It includes opinions, ideas, rumors, economic projections, statements of management’s future plans, and market commentary. The fact that hard information is quantitative means that it can easily be collected, stored, and transmitted electronically. This is why the advent of computers, large database programs, and networking has generated such growth in the use of production technologies that rely on hard information (e.g., quantitative lending, quantitative trading, and FinTech more generally).
- The unimportance of context. One can always create a numerical score from soft information, for example by creating an index of how honest a potential borrower is. This in and of itself doesn’t make the information hard. Your interpretation of a 3 must be the same as mine. Thus, a second dimension of hard information is the unimportance of the context under which the information is collected. One can collect and code information and then transmit it to someone else. The meaning of the information depends only upon the information that is sent. It does not depend upon dimensions of the environment under which it was collected but which are not encoded in the data (Ijiri 1975). Thus, the receiver of the data knows all that the sender knows (or at least all that is relevant). With soft
(^1) Text files can obviously be translated into numbers; this is how they are stored and transmitted. Can’t text files be processed electronically? Again, the answer has to be yes, conditional on what one means by processed. The abilityof computer algorithms to process and generate speech (text) has improved dramatically since we first discussed soft and hard information. Whether it can be interpreted and coded into a numeric score (or scores) is a more difficultquestion. A numeric score can always be created, the question is how much valuable information is lost in the process. We call this process the hardening of information and we will discuss it below.
1995; Aghion and Tirole 1997; Baker, Gibbons, and Murphy 2002). For a signal to be verifiable, the interpretation of the signal by the two contracting parties—and any third party who may be required to enforce the contract—must be the same. This is a characteristic of hard information. By contrast, soft information is private and not verifiable as it involves a personal assessment and depends upon its context, neither of which can be easily captured and communicated. Previous lenders can produce records showing that a borrower has paid their bills on time (hard information), but they cannot fully document for an unknown third party that a borrower is honest as this relies on multi-dimensional observations and on each party’s personal assessment and standards. Following the organizational economics literature, hard and soft information can also be referred to as objective or subjective information.
- Separation of information collection and decision-making. The unimportance of context for hard information means it is possible to separate the collection of hard information from the decision-making based on that information. Adam Smith observed that the division of labor and specialization can create value in manufacturing; the same principal applies to the collection and processing of hard information. Knowing what information to look for and why it is valuable (i.e., what will it be used for) is essential if information collection is to be delegated. The collection of hard information does not even need to be personal. Hard information can be entered into a form without the assistance of or significant guidance from a human data collector (home mortgage mobile apps are an example). The data collector does not need to understand what decisions the data will be applied to. Soft information must be collected in person, and the information collector and the decision maker are often the same person.^4 This
(^4) A typical example is the relationship-based loan officer. The loan officer has a history with the borrower and, based on a multitude of personal contacts, has built up an impression of the borrower’s honesty, creditworthiness, andlikelihood of defaulting. Based on this view of the borrower and the loan officer’s experience, the loan is approved or
is the intuition behind Stein’s (2002) argument that smaller, less hierarchical firms are better able to use soft information in their decisions. This is also why relationship lending is built upon soft information (Berger and Udell 1995). B) History of Soft and Hard Information Historically, most information collection and decision-making was local and between individuals who were familiar with each other, and thus the distinction between hard and soft information was not relevant. As technology made it feasible for financial transactions to occur between more distant and less familiar participants, the distinctions that we have been discussing started to arise. We have implicitly been assuming that information type is static. Information is either hard or soft; it is not malleable. This simplification allows us to focus on the definition and advantages of each type of information. How discrete and immutable is information type? That is an empirical question. We can think of hard information as a numeric index, but soft information can and is converted into an index, though not without a loss of information or context. Markets and individuals are constantly taking in soft (and hard) information and condensing (hardening) it into binary decisions: whether to fund a project, sell a stock, or make a loan. This does not create a meaningful loss of information if the decision is the final step and does not feed into later decisions.^5 Before moving on to the relative advantages of hard and soft information, we will discuss two historic examples of the hardening of information and the ways in which the process changed
denied. Uzzi and Lancaster (2003) provide detailed descriptions of such interaction between borrowers and loanofficers. (^5) In Bikhchandani, Hirshleifer, and Welch’s (1992) study of informational cascades, they model sequential decisions where agents see the (binary) decisions of prior agents but not the information upon which the decision is made. Thisreduction (hardening) of information leads to agents ignoring their own (soft) information and following the crowd.
agencies used this information to create two credit scores which were sold to merchants: pecuniary strength (essentially net worth) and general credit (ability and willingness to repay, Carruthers and Cohen, 2010b). In this way, the agencies were able to take the soft information based on personal contacts and available to local merchants and provide it in a form that was useful to distant merchants. Merchants could make lending decisions based on this number, even though they had no contact with the potential customers or the data collectors. The standardization of this information in the form of credit score resulted in a very early form of hard information, which allowed the geographic reach of trade credit lenders to expand. This is an example of how soft information can be hardened.
- Creation of Center for Research in Security Prices (CRSP). The second historical example comes from the equity markets. The Center for Research in Securities Prices began as a database of monthly and then daily returns on all NYSE stocks in the early sixties: stereotypic hard information. There is rarely disagreement about what a return of four percent means. Prior to the construction of the CRSP databases, however, there was limited knowledge about what the returns on equities actually were, let alone what the determinants of equity returns were. 9 The existence of a comprehensive database containing the returns on all stocks unleashed a torrent of research into the determinants of both expected returns (e.g., factor
(^9) CRSP began with a question from Louis Engel, a vice president at Merrill Lynch, Pierce, Fenner and Smith. He wanted to know what the long-run return on equities was. He turned to Professor Jim Lorie at the University of Chicago,who didn’t know either but was willing to find out for them (for a $50,000 grant). The process of finding out led to the creation of the CRSP stock return database. The fact that neither investment professionals nor academic financeknew the answer to this question illustrates how far we have come in depending upon hard information such as stock returns. Professor Lorie described the state of research prior to CRSP in his 1965 Philadelphia address: “Until recentlyalmost all of this work was by persons who knew a great deal about the stock market and very little about statistics. While this combination of knowledge and ignorance is not so likely to be sterile as the reverse—that is, statisticalsophistication coupled with ignorance of the field of application—it nevertheless failed to produce much of value.” In addition to CRSP, he talks about another new dataset: Compustat (sold by the Standard Statistics Corporation) whichhad 60 variables from the firm’s income statement and balance sheet. http://www.crsp.com/research/james-lorie-recognized-importance-crsp-future-research
models) and realized returns (e.g., event studies). It was now possible to carefully document what announcements or events influence stock prices (MacKinlay 1997). The dependent variable is a unidimensional index of value: the stock price (or changes in the stock price). The independent variables in this work are also coded into numeric values. Initially the coding was rudimentary: dividends increased, decreased, or did not change. Over time, the independent variables used to explain stock returns in event studies became more elaborate. However, they were always quantitative simplifications of the underlying events. Although the event studies often found important determinants of stock prices, even when they focused on the individual days when seemingly large announcements were made, the fraction of cross-sectional variability that the models were able to explain was small (Roll 1984; Roll 1988). This omission could be due to daily movements driven by the trading process (market micro- structure effects) or by the inadequacy of the explanatory variables. There are many forces that move stock price (e.g., rumors, news accounts, or different interpretations of public releases) that are not easily and accurately converted to a numeric score. Although market participants capture this soft information and impound it into the hard information of stock prices, the academic models have had difficulty replicating the process.
III) Advantages and Disadvantages of Hard Information. The choice between hard and soft information is driven by its availability and, more importantly, by the relative costs of each. In this section, we describe the relative advantages of hard or soft information. The objective is both to explain why one kind of information is preferable in a given context, but also to more fully understand the definition of each. A) Lower Costs of Production and Market Competition.
loans or services. In addition to expanding the number of suppliers in a given market, a reliance on hard information can also increase the geographic reach and competitive impact on existing suppliers. The evolution of the mortgage and signature loan (now called the credit card) market is an example.^12 In the 1950s, the market was local and based on soft information obtained through personal contact. It is now national and based on hard information often obtained through impersonal contact. This has led to a wider availability of and arguably cheaper capital for the middle class (Nocera 2013). The nature of the information may also increase the competitiveness of the markets. Once information is systematized and easy to communicate (hard), it also becomes more difficult to contain. In the early years of the credit reporting agencies (e.g., J. M. Bradstreet & Son or R. G. Dun), only a summary of the information the agencies had on borrowers was published in their quarterly books. This disclosed information was quantitative and easy to compare and communicate. For an additional fee, subscribers could visit the office of the agencies to view a detailed report on a potential customer. The credit rating bureaus were either unable or unwilling to quantify and include all of the soft information they held into their reported credit scores. Interestingly, information in these private reports was better at predicting bad outcomes (business failures) than the published credit ratings (Carruthers and Cohen 2010b). By keeping the information difficult to replicate and transmit, by maintaining its softness, the credit reporting agencies hoped to maintain their control over the information and thus extract greater rents from the information they had collected. Once information is hard, providers have difficulty preventing one customer from passing it to additional customers who can then capture the information’s full value. Information that is hard can be understood independent of the collector and the context
(^12) Subprime mortgage loans are less standardized and more informationally sensitive than normal mortgages because sometimes borrowers are not able to provide full disclosure of their income (Mayer, Pence, and Sherlund 2009).
under which it was collected. If the collector is not necessary once the user has the data, this makes charging high rates for the information more difficult. B) Durability of Information. The durability of information is also greater when it is hard. The fact that it is easily stored means that the cost of maintaining it for future decisions is low. The fact that the information can be interpreted without context means that it is possible to pass it along to individuals in different parts of an organization (Stein 2002). Individuals or even firms no longer need to be part of the data collection process to be part of the decision-making process. This ease of interpretation is especially important if the people involved in data collection are not expected to be around in the future. It effectively frees the decision process from constraints of space (distance) and time. Given the increased turnover in many finance professions (loan officers or investment bankers), the movement toward hard information seems inevitable.^13 As described in Crane and Eccles (1988), junior investment bankers used to rise through the bank at the same time as junior employees of their clients were rising through the ranks at their own firms. By the time junior bankers became senior bankers, they had developed a relationship with the people who were now in senior management positions at the client firms. There is no need to rely on formal records (hard information) in the presence of these long-term relationships. However, if bankers turn over more frequently, new bankers must rely on the records left behind by the previous bankers (Morrison and Wilhelm 2007). This creates a greater reliance on hard information. C) Lost Information.
(^13) Karolyi (2017) finds that the relationship lies with the individuals, not the firms. After exogenous changes in leadership (the death or retirement of a CEO), firms are significantly more likely to switch to lenders with whom thenew CEO has a relationship (see also Degryse, Liberti, Mosk, and Ongena 2013). This is one reason why firms that rely on soft information in securing debt capital care about the fragility of the banks from which they borrow (Schwert2017).
decisions is driven by this question (O’Neil 2016). If there are borrowers that are really good credit risks, but they look bad on paper (i.e., when only hard information is considered), then such borrowers would be incorrectly denied credit. How often are such mistakes happening? The empirical evidence thus far is mixed. It is clear that some small borrowers are dislocated by their banks when the banks merge, but there is also evidence that existing and new small banks may fill the gap (Berger, Miller, Petersen, Rajan, and Stein 2005; Berger, Goulding, and Rice 2014; DeYoung, Gron, Torna, and Winton 2015; Berger, Bouwman, and Kim 2017). D) Gaming the System. Accounting numbers, such as a firm’s income statement and balance sheet, are a classic example of hard information. The information is all quantitative, it is easy to store and transmit electronically, and there is relatively uniform agreement about what numbers like revenues and costs mean. Quantitative decisions from asset allocation to credit approval all rely on these numbers because of those characteristics. At the same time, newspaper accounts of accounting manipulation and the size of the credit rating manuals make it clear that these decisions are not simply a function of the numbers the firms disclose. This ambiguity raises another cost of using hard information (e.g., automated or delegated decisions methods): a loss of certainty regarding who controls the information which is fed into the decision-making process. The discussion thus far has focused on the decision maker (e.g., the loan officer making a loan decision), not the target of the decisions (e.g., a loan applicant). By choosing to use hard versus soft information, the (ultimate) decision maker is trading off the lower cost of collecting and processing the information with potential loss in accuracy of the information upon which they are basing their decisions. The way a decision is made, including the type of information upon which the decision depends, will also influence the actions of the target of the decision.
The behavioral response of borrowers (or other targets of the decision) places restrictions on how decisions based on hard information can be made. Having a decision depend entirely upon the numbers and a transparent decision rule can work, but only if the cost of manipulating the numbers is sufficiently high relative to the benefit of the preferred outcome. 16 If a firm can raise its reported assets or sales by a small amount for a small cost, and this will raise its credit rating and lower its cost of capital sufficiently, it has an incentive to inflate its reported assets or sales.^17 The rules cannot be a direct and transparent function of the hard numbers if the hard numbers are under the discretionary control of the target of the decision. In this case, the decision maker has an incentive to make the decision a fuzzy and opaque function of the inputs. The line between an AA and an A rating can be kept secret or additional sources of soft information can be included.^18 In
(^16) In the financial crisis of 2008, a large number of investment grade securities defaulted. The magnitude of the defaults suggested there was a problem with the rating process (see Benmelech and Dlugosz 2009a; Benmelech and Dlugosz2009b). Observers in industry, academics, and government suggested possible sources of the problem and potential solutions. What is intriguing is the defaults experience was very different in the corporate bond market (debt ofoperating companies) compared to the structured finance market (e.g., RMBS). Defaults in the corporate bond market spiked in 2009, but the peak is not drastically different than the peak in prior recessions (see Vazza and Kraemer 2016,Chart 1). The peak in defaults in the structured finance in 2009 is dramatically larger (see South and Gurwitz 2015, Chart 1). Although the collapse of the housing market hit the structured finance securities harder, this suggests that apart of the problem with the rating process resides uniquely in the structured finance segment of the market. For an operating company, a low cost of capital is an advantage but not its only or predominant source of competitiveadvantage. For a securitization structure, a lower cost of capital is one of its few source of “competitive advantage.” Thus, a bank might change which mortgages are placed into a securitization if this change would increase the factionof the securitization rated AAA and thus lower the cost of capital. An auto-manufacturing firm is unlikely to close plants or close down a division solely to get a higher credit rating. The costs of altering the business to improve acredit score are higher and the benefits are (relatively) lower for an operating firm. This may be why we saw relatively fewer defaults in the corporate bond sector relative to the securitized sector. This issue prompted the credit ratingagencies to consider different rating scales for structured finance versus corporate debt (Kimball and Cantor 2008). (^17) Hu, Huang, and Simonov (2017) see the same behavior in the market for individual loans. The theoretical importance of nonlinearities in the mapping of inputs (hard information) to outputs (decisions) is discussed in Jensen(2003). In his examples, the incentives to misstate one’s information are smaller if the payoff function is linear. Small changes in the reported information have only small changes in the manager’s payoff. (^18) There may also be strategic reasons to avoid a transparent mapping between the numbers and the credit rating. The business model of credit rating agencies relies on market participants being unable to replicate the ratings at lowercost than the agency. If the mapping were a direct function of easily accessible inputs (e.g., the income statement and balance sheet) and nothing else, some clever assistant finance or accounting professor would figure out the function.This is one reason that the early credit reporting agencies released only a fraction of their information publicly in the form of a credit score. For additional fees, users could review a more complete report (Carruthers and Cohen 2010a,Cohen and Carruthers 2014).
lending their own capital, but the bank’s. The bank manager or shareholder must trade-off the value of the loan officer using their soft information (better quality decision and lower transactions costs) against the misaligned incentives between the loan officer and the bank. The advantage of hard information is that it can remove the loan officer’s discretion. The relevant variables and the mapping from the variables to the decision is beyond the control of the loan officer in these cases.^21
IV) Traditional Banking and the Organizational Design of Lending. The evolution of financial markets over the past forty years has been in part a replacement of soft information with hard information as the basis for financial transactions. The full ramifications of this transformation are not yet fully apparent, and as we discussed above, there are both advantages and disadvantages of this transformation. In this section, we describe the evolution of soft information since its theoretical origins, the application of the concept of information type in the traditional banking literature, and its implications for the organizational design of lending by financial intermediaries. A) Beginnings: Theoretical Literature. The finance literature has been exploring the distinction between soft and hard information for several decades now, and our understanding has evolved since the early years. The distinction was not always explicitly stated, and even when it has been, the definition was not complete,
inside a German savings bank but find no evidence that loans approved based on discretion perform differently thanthose that do not use discretion. Cerqueiro, Degryse, and Ongena (2011) find that discretion seems to be important in the pricing of loans, but plays only a minor role in the decision to lend. (^21) This turns out to be an imperfect solution when the loan officer has an incentive and the ability to manipulate the inputs, just as the borrower might. The loan officers in Berg, Puri, and Rocholl (2016) work for a bank that uses aninternal credit score to evaluate loans. They show that loan officers repeatedly enter new values of the variables into the system until a loan is approved. Not only are they able to get loans approved that were originally rejected, but theyalso learn what the model’s cut offs are and thus what is required to get a loan approved. These results suggest that even hard information decision-making algorithms which are transparent and depend upon data subject to the controlof either participant (local decision maker or the target of the decision) are subject to the Lucas critique (see the Gaming the System discussion above).
formally treated, or consistent across applications. One origin of soft and hard information traces back to the theoretical financial intermediation literature and the distinction it drew between the role of banks (or other private lenders) versus the public bond markets. A key distinction was the superior ability of banks to collect and process information (Diamond 1984; Diamond 1991; Ramakrishnan and Thakor 1984; Allen, Carletti, and Gu 2015). This explained why many opaque firms relied exclusively on banks. The public debt markets, however, with the help of rating agencies, have the same job description: to evaluate the credit quality of firms (Ederington and Goh 1998). The difference is the type of information each specializes in collecting and processing. The public bond markets and the rating agencies collect financial disclosures, accounting reports, and default histories. These are sources of hard information. They can all be reduced to a series of numbers. Banks, on the other hand, especially as described by the lending relationship literature, collect information that is neither initially available in hard numbers (the ability of the managers, their honesty, the way they react under pressure), nor easily or accurately reducible to a numerical score. Even if reduced to a numerical score, the interpretation of the information may be judgmental and include a discretionary component (Cole, Goldberg, and White 2004; Hertzberg, Liberti, and Paravisini 2010). Once the relationship is established, even then this information is not hard. The firm is still unable to communicate this information to the broader lending markets and thus negotiate a lower loan rate from its bank (Petersen and Rajan 1994). Originally, finance scholars borrowed the concept of soft information from organizational economics and the theoretical literature on decision making in organizations. One feature of those initial models was that the interests of the parties were imperfectly aligned. This misalignment created incentives for individuals to distort the information that was collected and transmitted in a
