E VALUATING THE IMPACT OF M OROCCAN COMPANY LAW REFORM ON MANUFACTURING FIRMS ’ ACCESS TO CREDIT
A thesis submitted in partial fulfillment of the requirements for the degree of M.P HIL IN E CONOMICS by
S IMON Q UINN LLB(Hons) BEcon(Hons) Qld
K EBLE C OLLEGE , U NIVERSITY OF OXFORD .
T RINITY 2007
Supervisor: Professor Marcel Fafchamps
ABSTRACT
In 2001, Morocco introduced a radical new company law regime. This thesis evaluates the impact of that change upon manufacturing firms’ access to credit. The thesis develops a signalling model in which a representative bank conditions its credit decision upon a representative firm’s choice of legal status; legal status acts in the model as an informative proxy for characteristics unobservable to the bank. The model shows that choice of legal status may be significant in determining provision of bank credit, both because of the substantive obligations that it imposes upon firms (incentive effects) and because of its role as a signal of hidden information (information effects). The thesis tests these insights using panel data. Legal status is found to be significant in explaining provision of overdraft facilities, and it is argued that the relationship is causal. Consequently, legal status is also significant for the size of the overdraft limit. However, status does not significantly explain quantity of bank lending, either for short- or long-term debt. Finally, the thesis proposes and applies a new methodology for testing for information effects — distinct from incentive effects — in a panel data context; the method involves testing for a non-monotonic relationship between firm quality and change in credit outcome. The methodology finds some evidence that legal status has a significant information effect, but is unable to reject the null hypothesis of perfect information.
ACKNOWLEDGEMENTS
Most importantly, I must thank my supervisor, Professor Marcel Fafchamps. Professor Fafchamps has been a thoroughly invaluable source of ideas and suggestions throughout this research. I have been fortunate to have learned many specific techniques from working with him — but, even more than learning any particular skill, I have appreciated most the opportunity to learn from Professor Fafchamps many general insights into the process of economic research. It has been a privilege to work with him. I took the opportunity during last summer to visit Morocco, to speak with policymakers and members of the business community. I found the visit exceptionally useful, both in framing specific research questions and in generally informing my sense of the issues facing the Moroccan economy. I must acknowledge funding for the trip from the George Webb Medley Fund (Department of Economics) and the Keble Academic Fund, in addition to the general funding generously provided to me throughout my studies in Oxford by the Rhodes Trust. I must thank all of those people who took the time to speak with me in Morocco: Nadia Amrani (USAID), Najy Benhassine (The World Bank), Mehdi Bouziane and Abdallah Chater (Centre R´egional d’Investissement, Casablanca), Richard Cantin (Naciri & Associ´es, lawyers), Yassir Charafi (The International Finance Corporation), Carl Dawson (the American Chamber of Commerce in Morocco), William Fellows (Financial Services Volunteer Corps), senior representatives of Bank al Aman, Banque Centrale Populaire and the BMCE Bank, and several other representatives of the Moroccan business community. Similarly, I thank Fraser Thompson, both for his general guidance on researching Morocco and for his specific help in planning my visit. I emphasise that this research does not necessarily represent the opinions of any of the people with whom I spoke; nonetheless, their time and insights have proven essential. Finally, I must acknowledge my peers who have helped me so much with so many important questions, primarily relating to problems with STATA and LATEX but extending even to queries about maths and about corporate law; many thanks to Angela Ambroz, Henry Ashton, Andrew Marshall, Justin Sandefur, Nick Woolley and Andrew Zeitlin (Oxford University) and to Joseph Clark (The University of Queensland). For their help with my translations, thanks to Thomas Flury and Clotilde Giner. Of course, ‘the usual caveat’ applies, and I do not mean to implicate anyone else in the shortcomings of the present research. Nonetheless, it is no exaggeration that this research has only been made possible by the generous assistance of many people; I am extremely grateful for their help.
WORD COUNT, SOFTWARE AND STYLE
Word count A straight automated word count, including all typesetting commands, produces a result of approximately 26 500 words. Alternatively, an estimation of 290 words per page × 100 substantive pages suggests a length of 29 000 words. Note that the Appendix has been excluded from the word count. It is a summary of the key aspects of the legal reform, and is provided only for the reader’s convenience; to the author’s knowledge, this appendix is the only summary of the reform available in English.
Software Econometric analysis for this thesis was conducted with STATA (version 9.2), including the kernel regression program kreg.ado of Fafchamps (2002). Diagrams were created in CorelDRAW 9 and Microsoft Office Excel 2003.
Style The thesis is typeset in LATEX 2ε and, for what the reader may comfortably assume are reasons of style and convenience rather than of unbounded optimism, referencing follows the style of the journal Econometrica. Spelling and grammar follow the British conventions.
CONTENTS
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
2. The context: Morocco’s legal reform . . . . . . . . . . . . . . . . . . . . . 2.1 The process of legal reform . . . . . . . . . . . . . . . . . . . . . . . 2.2 Comparison: SA and SARL companies under the new law . . . . . . 2.3 Comparison: SA obligations before and after the legal change . . . . 2.4 A migration away from the SA status . . . . . . . . . . . . . . . . .
5 6 9 9 10
3. Theoretical model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Motivating literature: Credit constraints and information . . . . . . . 3.2 Conceptual framework . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Modelling a single, unobserved, continuous type . . . . . . . . . . . 3.3.1 Developing the model . . . . . . . . . . . . . . . . . . . . . 3.3.2 Solving the model . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 Insights from the univariate model . . . . . . . . . . . . . . . 3.4 The bivariate case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Developing the bivariate model . . . . . . . . . . . . . . . . 3.4.2 Solving the model . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 Applying the bivariate model . . . . . . . . . . . . . . . . . . 3.5 Conclusions and testable hypotheses . . . . . . . . . . . . . . . . . .
13 13 15 16 16 20 34 34 35 36 37 47
4. Testing strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Outline of the testing strategy . . . . . . . . . . . . . . . . . . . . . . 4.2 Conceptualising the legal reform as a ‘treatment’ . . . . . . . . . . . 4.3 Fixed-effect estimators . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Identifying asymmetric information by non-monotonicity . . . . . . . 4.5.1 Summary of related literature . . . . . . . . . . . . . . . . . 4.5.2 Identification through non-monotonicity . . . . . . . . . . . . 4.5.3 Identifying a concave relationship . . . . . . . . . . . . . . . 4.5.4 Significance testing . . . . . . . . . . . . . . . . . . . . . . . 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51 51 53 55 56 61 62 63 66 67 69
5. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Data sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . .
70 70 71
6. Econometric results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Hypothesis 1: Choosing SA rather than SARL causes banks to be more willing to provide credit . . . . . . . . . . . . . . . . . . . . . 6.1.1 Hypothesis 1.1: Concerning provision of an overdraft facility 6.1.2 Hypothesis 1.2: Concerning limits on the overdraft facilitiy . . 6.1.3 Hypothesis 1.3: Concerning short-term lending . . . . . . . . 6.1.4 Hypothesis 1.4: Concerning long-term lending . . . . . . . . 6.2 Hypothesis 2: Signalling effects contribute to the relative value of choosing SA status . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Empirical conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .
91 95
7. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
96
74 74 74 80 83 88
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 8. Appendix: Details of the legal reform . . . . . . . . . . . . . . . . . . . . 8.1 An overview of Moroccan law . . . . . . . . . . . . . . . . . . . . . 8.2 The new legal regime: a comparison between SA and SARL . . . . . 8.3 The old and the new: a summary of the legal reform . . . . . . . . . .
i i ii iv
LIST OF TABLES
2.1 2.2 2.3 2.4 2.5
SA and SARL: Key legal differences . . . . . . . . . . . . Key legal reforms: the SA form . . . . . . . . . . . . . . . Summary of aggregate migration, 1997-2003 . . . . . . . . Aggregate change in legal status, FACS and ICA surveys . . Details of movement in legal status, FACS and ICA surveys .
. . . . .
9 10 11 11 12
5.1 5.2
Summary of variables: Two-period panel . . . . . . . . . . . . . . . . Summary of variables: Five- and six-period panel . . . . . . . . . . .
72 73
6.1 6.2 6.3 6.4 6.5
Fixed-effect logit: Impact of legal status upon bank overdraft . . . . . Fixed-effect OLS: Instrumenting for legal status . . . . . . . . . . . . Fixed-effect OLS: Impact of legal status upon bank overdraft limit . . Difference in mean overdraft across migration categories . . . . . . . Fixed-effect OLS: Impact of legal status upon short-term debt, FACS & ICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . First-differences OLS: Impact of legal status upon short-term debt, 1997-2002 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed-effect OLS: Impact of legal status upon long-term debt, FACS & ICA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . First-differences OLS: Impact of legal status upon long-term debt, 19972002 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simulation results: 1000 valid simulations at varying bandwidths . . .
75 77 81 83
6.6 6.7 6.8 6.9
. . . . .
. . . . .
. . . . .
. . . . .
85 87 89 90 93
LIST OF FIGURES
2.1 2.2
The introduction of Law 17/95 . . . . . . . . . . . . . . . . . . . . . Migration from the SA to SARL status: The census data . . . . . . .
8 11
3.1 3.2 3.3 3.4 3.5
19 40 40 41
3.8
The structure of the game . . . . . . . . . . . . . . . . . . . . . . . . The ‘firm indifference curve’ . . . . . . . . . . . . . . . . . . . . . . Drawing inference in the bivariate normal case . . . . . . . . . . . . The ‘bank indifference curve’ . . . . . . . . . . . . . . . . . . . . . . Separating and pooling equilibria when the bank indifference curve is steeper than the firm indifference curve . . . . . . . . . . . . . . . . Separating and pooling equilibria: Credit constraints and non-performing loans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A separating equilibrium when the bank indifference curve is flatter than the firm indifference curve . . . . . . . . . . . . . . . . . . . . . A shift in the firm indifference curve . . . . . . . . . . . . . . . . . .
45 46
4.1
A non-monotonic loss in the bivariate normal case . . . . . . . . . . .
65
6.1 6.2 6.3
Estimated relative effect of choosing the SA status . . . . . . . . . . Kernel regression of those firms switching legal status (Bandwidth: 1.1) Kernel regression of those firms switching legal status (Bandwidth: 1.1): Focusing upon the turning point. (Points plotted are the kernel estimate points, not the data points.) . . . . . . . . . . . . . . . . . . Imputed p-values across varying bandwidth . . . . . . . . . . . . . .
79 92
3.6 3.7
6.4
42 43
92 94
1. INTRODUCTION
Motivation Legal reform is often viewed as an important component of a nation’s strategy for development. However, it is a component whose effects — even immediate effects — are often very difficult to assess accurately. Panel data sets bridging such reforms are scarce, and the kind of transformational legal reform likely to reveal the most significant insights into economic behaviour is unusual, particularly in the context of developing economies.
This thesis attempts such an evaluation. It considers the impact of a radical overhaul of Moroccan company law upon the relationship between Moroccan banks and the manufacturing sector. Specifically, it explores the effect of firms’ choice of legal status on credit market outcomes. Moroccan firms may choose to operate as a Soci´et´e Anonyme (SA) or under the less onerous form of Soci´et´e A` Responsibilit´e Limit´ee (SARL); this thesis considers the effect of the new Moroccan company law upon that choice, and the consequent impact upon manufacturing firms’ access to credit. In doing so, the research responds to several motivations. First, by a quirk of fortune (outlined in Chapter Two), a delay to the implementation of Morocco’s new company law regime allowed two rounds of a panel survey of Moroccan manufacturers to form a ‘before/after’ snapshot of the introduction of the reform. This, in itself, is a significant motivation: it means that, to the author’s knowledge, this research is the first specific econometric evaluation of the new Moroccan law, which itself forms an important example of a
1. Introduction
2
significant legal modernisation in an emerging economy. Second, such an evaluation may potentially provide valuable contribution to the debate in Morocco. As Chapter Two will show, Morocco’s new company law provoked significant controversy and discussion in Morocco; debate and lobbying on further reforms is presently ongoing. Finally, the research considers an issue often identified as significant in the context of developing economies: profitable firms being unable to obtain bank credit as a result of asymmetric information problems. As an emerging economy with a developing banking sector, Morocco is a useful environment in which to consider this problem; moreover, considering the relationship between credit constraints and Morocco’s legal reform draws particular attention to the respective roles of corporate governance and of asymmetric information in developing economies’ credit markets.
Contribution This thesis seeks to contribute to the literature in three respects: theoretical, empirical and methodological. Theoretically, the thesis develops a new model of credit constraints under asymmetric information. As Chapter Three shows, this is an issue that has been explored in the theoretical literature before. The contribution of this model is to apply a standard signalling framework to a situation where firm quality is partially observed by banks, and where legal status may therefore act as an informative signal of unobserved firm characteristics. This application illustrates that legal status may indeed have a signalling effect. Further, the model shows that the informativeness of that signal will depend upon its cost in terms of unobserved firm characteristics, relative to observed characteristics. Finally, the model provides a simple way of conceptualising the effects of a legal reform: changes in the cost of signalling induce changes in firms’ legal status which, in turn, induces changes in their access to credit. The contribution of this model is thus to provide a straightforward and intuitive mechanism for considering the role of a binary signal in a context of partially asymmetric information.
1. Introduction
3
Empirically, the thesis conducts the first microeconometric work focussed specifically upon the impact of Morocco’s new legal regime. Empirical results show that legal status has a significant impact both upon whether a firm has a bank overdraft and upon the limit of that overdraft or upon other bank lending. This is important for showing that, in some respects at least, Moroccan banks use legal status as an important factor in their lending decisions. In turn, this is an important indication that, by increasing the obligations upon SA companies, Morocco’s legal reform is likely to have significant ongoing consequences for Moroccan credit markets.
Methodologically, the thesis proposes a new way of distinguishing information effects from incentive effects in the empirical analysis of contracting behaviour. Chapter Four shows that this has been an ongoing empirical challenge for literature in this area. Further, that chapter shows that two bases for distinction have been used: differentiation between classes of contracts and differentiation between times in the contracting process. This thesis suggests a third basis for distinction: differentiation across a continuum of agent types. The thesis explains how simulating the perfect-information null hypothesis with kernel regression can be used to test for significance in this context. When the methodology is applied to the present data, we find some evidence of asymmetric information effects — distinct from incentive effects — but cannot reject the null hypothesis of perfect information.
Outline The thesis proceeds as follows. Chapter Two sets out the background to the work. It summarises both the process of Morocco’s corporate reform and the key aspects of the new legislation. The purpose here is not jurisprudential; rather, the section is concerned to show (i) that the new legislation fundamentally altered Moroccan company
1. Introduction
4
law (so that the relative cost of the SA form increased substantially), and (ii) that, despite a long process of drafting and debate, the final (and crucial) implementation of the legislation fell between the two rounds of the relevant panel survey. Chapter Three develops a theoretical model, by reviewing and developing existing theoretical literature. The model is developed for the purpose of presenting an internally-consistent hypothesis for empirical analysis. Specifically, the model is a ‘signalling model’, where legal status plays a role as an signal of unobservable firm quality. The model is developed in two steps: first, as a simple univariate model (where firm quality is completely unobservable to banks) and then, building upon this, as a bivariate model (where firm quality comprises both an observed and an unobserved component). The model illustrates that, in theory at least, choice of legal status can play an important signalling role, and that this effect may differ across observable covariates. Specific testable hypotheses emerge.
Chapter Four outlines the strategy for testing those hypotheses. The chapter conceptualises the legal reform as a ‘treatment’, and explains concerns for endogeneity in this context. Appropriate estimators are outlined. Further, robustness checks are suggested, including instrumenting. Instrumenting poses particular methodological problems in the context of a binary dependent variable, and a response to this is proposed. Finally, the chapter outlines the new methodology for testing for asymmetric information through non-monotonicity. Chapter Five outlines and summarises the data available; Chapter Six applies to that data the strategies outlined in Chapter Four. The predictions of the model are confirmed in the case of provision and limits of bank overdraft, but the no-effect null hypothesis cannot be rejected in considering bank lending. The chapter concludes by applying the new methodology to test for asymmetric information; the methodology finds evidence of asymmetric information, but not of a sufficient significance to reject a perfect-information hypothesis. Chapter Seven concludes.
2. THE CONTEXT: MOROCCO’S LEGAL REFORM
This chapter summarises both the process and the substance of Morocco’s introduction of a new company law regime. In doing so, it seeks three aims. First, it identifies the key dates for the reform. This is important for confirming that the reform occurred between the FACS survey in 2000 and the ICA survey in 2004 (discussed in Chapter Five); the analysis shows that, thanks to delays, the crucial date of final implementation was 1 January 2001. Second, the summary compares the SA and SARL legal forms under the new legal regime, to show that the SA form indeed imposes substantially more onerous obligations, and that these obligations are likely to prove relatively more costly for firms of lesser quality. This insight is fundamental to motivating the theoretical model developed in Chapter Three, because — as that chapter will show — the assumption that the relative cost of the SA status is negatively correlated with firm quality is central to justifying a signalling model. Finally, the section compares the old legal regime with the new. In doing so, it shows that SA firms faced substantially greater obligations after the legal reform than before. This substantial before/after difference is central to motivating the entire research: it explains why the reform is justifiably viewed as a ‘quasi-experiment’ in the Moroccan credit market and why, as summary statistics in this chapter will illustrate, a large number of firms abandoned the SA status in response to the legal change.
2. The context: Morocco’s legal reform
2.1
6
The process of legal reform
From 11 August 1922 until 20 January 1997, Moroccan SA companies were governed exclusively by French law; a dahir of 1922 applied to Morocco the relevant provisions of the French commercial law of 24 July 1867 (Lazrak 2004, Slaoui and Lecerf 2000, Chbani Idrissi 1996). That law reflected very early notions of corporate governance and accountability; a prominent French lawyer commented in 2005 that the regime allowed an “anarchy” (L’Economiste 2005), while one Moroccan businessman interviewed for the present research described the former legal regime as “a jungle”.1 SARL companies were governed by a regime of the same era: a dahir implemented in September 1926 (Lofti 1996).
However, as the end of the last century neared, Moroccan legislators came to recognise the need for a new corporate law regime to reflect and support the modernisation both of Moroccan law and the Moroccan economy. Thus, on 2 July 1996, the Chambre des Repr´esentants adopted Law 17-95: a new regime to regulate SA companies.2 The authors (two French lawyers) took as their starting point the French company law of 24 July 1966 (L’Economiste 1999b, Lazrak 2004). Law 17-95 was promulgated by dahir on 30 August 1996.3 However, rather than being immediately applicable from that date, the law created a dual regime for its implementation. New SA companies would be required to have corporate statutes in compliance with the law if created after the entry into force of relevant provisions relating to the register of commerce (Article 443, Law 17-95); SA companies created prior to that date would receive an additional grace period to harmonise their corporate statutes with the law (Article 444, Law 1795). Progress of the new SARL law followed a similar trajectory. A statute to replace the 1926 law was adopted on 7 January 1997 and promulgated by dahir on 13 February 1 2 3
Interview with the author, Casablanca, September 2006. (Name withheld from publication here.) Dahir no 1-96-124, preamble. Dahir no. 1-96-124. Dates and details cited are provided in the preamble to the dahir.
2. The context: Morocco’s legal reform
7
of that year. This law (Law 05-96) was subject to a similar dual regime (under Articles 121, 128 and 129 of that law).
The relevant provisions relating to the register of commerce were duly brought into force on 20 January 1997. Thus, law 17-95 was first applicable from this date, for new companies created in the SA status (Maatouk 2002, L’Economiste 1999c, L’Economiste 1998a).4 Law 05-96 would have been operative from the same time (Article 120, Law 05-96), but was not promulgated until 13 February; thus, it applied immediately from this later date. Identifying the required date of harmonisation for existing companies was much less straightforward. The relevant provisions of each law originally provided that the harmonisation of existing company statutes would occur within two years. However, there was confusion from the outset (apparently arising because of an imprecise translation from the official Arabic text to the French version) as to when the two year period would commence (L’Economiste 2000a, L’Economiste 1999c); as a consequence, interpretations differed between experts and even between government authorities in different cities (L’Economiste 1999a). Eventually, consensus settled upon the most generous interpretation; namely, that the dual-regime articles allowed for a grace period of two calendar years following the year that the statutes were first applicable (that is, 1997).
It was anticipated, then, that law 17-95 and law 05-96 would have universal application from 1 January 2000. However, with just three days remaining until the deadline, the parliament intervened by amending the relevant articles of both laws to extend the grace period for a further year (L’Economiste 2000b); the amendment was promulgated by dahir the following day.5 Despite calls for another similar amendment a year later, the 4
5
Apparently, existing SA companies were entitled to bring their corporate statutes into compliance with the law before this date; however, as subsequent discussion will explain, there were no sanctions for not doing so. Dahir no. 1-99-327 and no. 1-99-328 of 30 December 1999.
2. The context: Morocco’s legal reform
8
Fig. 2.1: The introduction of Law 17/95
grace period finally expired on 31 December 2000. Thus, from 20 January 1997 until 31 December 2000, new SA companies created were required to comply with law 1795, while existing SA companies could ‘opt-in’ to law 17-95 or remain under the old law (ie the dahir of 1922). The same principle applied to SARL companies. It was not until 1 January 2001, more than four years after its promulgation, that law 17-95 finally applied universally to all SA firms (and, similarly, law 05-96 to all SARL firms). Crucially — thanks to the additional delay — this implementation occurred after the completion of the FACS survey of manufacturing firms in 2000; for this reason, the FACS-ICA panel (summarised in Chapter Five) provides a before/after comparison of the effects of the implementation of the new legal regime.6 Figure 2.1 summarises the process of reform.
6
Article 449 of law 17-95 and Article 126 of law 05-96 provided the consequences of failing to harmonise the corporate statute: a fine of 2000 to 10 000 dirhams could be imposed upon the relevant individuals and a court-imposed grace period of less than six months allowed; if the subsequent grace period was not followed, an additional fine of 10 000 to 20 000 dirhams was allowed (in addition, presumably, to all of the other consequences specified for operating in breach of the relevant laws).
2. The context: Morocco’s legal reform
2.2
9
Comparison: SA and SARL companies under the new law
The appendix provides a detailed summary of the changes wrought by the new law, and their comparison to the previous legal regime. For now, we outline only the key points. Table 2.1 summarises the main differences between law 17-95 and law 05-96. It is clear, even from a brief overview, that the relative cost of adopting the SA form (rather than the SARL form) is significant. Further, it is equally clear that the relative cost is higher still for those firms of lesser quality; for example, the relative cost is higher for firms that are less structured, less transparent, less wealthy and in less need of equity finance. This point — that the relative cost of the SA form is negatively correlated with firm quality — will shortly prove fundamental in developing a signalling model of the reform. Tab. 2.1: SA and SARL: Key legal differences Soci´et´e Anonyme Soci´et´e A` Responsibilit´e Limit´ee Capital Minimum capital requirement Share transferability Granting security Governance Governance structure Transparency Appointment of auditors
2.3
300 000 or 3 000 000 MAD Freely transferrable May grant all forms
100 000 MAD Limited (subject to approval) Limits on moveable interests and negotiable securities
Dualist
Unitary
Must always have at least one
Generally need not have any
Comparison: SA obligations before and after the legal change
Though the reforms did include a new SARL law, the more substantial changes by far occurred to the SA status; it is here that legislators’ concerns about corporate governance were most clearly expressed, and here that the most controversy was raised. Table 2.2 summarises the key reforms to the SA status; it shows that the cost of operating as an SA company (relative to the cost of operating as an SARL company) was increased substantially by the new obligations of Law 17-95.
2. The context: Morocco’s legal reform
10
Tab. 2.2: Key legal reforms: the SA form Governance Management structure Limitations Management involvement Transparency Auditors Right to information Conflicts of interest Minority rights Procedural rights Third parties Risk of non-compliance with corporate statutes Corporate legal personality
2.4
Before
After
Unitary Only the corporate object Allowed ‘sleeping directors’
Dualist System of review Encourages active involvement
Played a relatively lesser role Almost none Clear conflicts allowed
Active and ongoing role Extensive and ongoing Prior and subsequent authorisation required
Almost none
May convene a general meeting
Borne by the third party Upon constitutive meeting
Borne by the company Upon registration
A migration away from the SA status
The introduction of the new laws provoked a marked migration away from the SA corporate status to other forms of corporate status (predominantly the SARL status); by imposing greater costs upon firms (that is, both direct compliance costs and the threat of sanction), the law made it relatively more attractive for firms to choose other corporate forms.7 However, the migration did not begin with the universal application of the law. Rather, as one would have anticipated, many companies abandoned the SA form in anticipation of the application of the new regime. Indeed, this was noted in L’Economiste as early as 1998 (L’Economiste 1998b).
Figure 2.2 shows the trend of the migration, recorded in the census data; it shows the number of firms changing from the SA to the SARL status by their year of first migration (that is, the year that the firm was first observed as being SARL status, having previously been SA).9 Table 2.3 summarises the aggregate transition as recorded in 7
9
For example, Azzedine Benmoussa (Founding Director of KPMG Morocco) noted in 2002 that “a large proportion of SA companies, essentially of the family type, transformed themselves into SARL with the entry into application of law 17-95 on the soci´et´e anonyme (SA)” (Shamamba 2002).8 This conclusion accorded closely with the responses of Moroccan businesspeople interviewed for this research; indeed, no other explanation for the migration was ever suggested. ‘First’ migration because, of the 1601 firms migrating SA → SARL in the census data, 19 are recorded — accurately or not — to have done so twice; that is, to have switched SA → SARL → SA → SARL.
2. The context: Morocco’s legal reform
11
the census; Tables 2.4 and 2.5 do the same for the panel.10 Fig. 2.2: Migration from the SA to SARL status: The census data
Tab. 2.3: Summary of aggregate migration, 1997-2003
Not changing status Switching SA → SARL Switching SARL → SA
Census, all firms Freq. Per. Cum. 9 018 84.0% 1 601 14.9% 98.9% 115 1.1% 100.0%
Census, panel firms Freq. Per. Cum. 216 42.3% 283 55.4% 97.7% 12 2.3% 100.0%
Tab. 2.4: Aggregate change in legal status, FACS and ICA surveys
Legal status SARL SA Total
10
FACS survey 324 188 512
ICA survey 415 97 512
Total 739 285 1024
Recall that the panel used includes only firms in either the SA or SARL status; thus, we would expect the same shape to the proportion of firms changing status over time, but would expect smaller percentages among the panel firms. The data reflects this.
2. The context: Morocco’s legal reform
12
Tab. 2.5: Details of movement in legal status, FACS and ICA surveys
Change, F ACS → ICA No change SA → SARL SARL → SA Total
Freq. 403 100 9 512
Per. 78.7% 19.5% 1.8% 100.0%
So three points emerge. First, the differences between the SA and SARL forms are substantial, and particularly pronounced for firms of ‘poorer quality’. Second, the reforms themselves were both radical and controversial, primarily for significantly increasing the direct costs of operating as an SA company — and this was reflected by a substantial induced migration away from the SA status. Finally, though the legal reform imposed differential costs, the decision as to whether a particular firm would stay as an SA or migrate to SARL remained a matter of the firm’s choice. And that, no matter how much legal theorists may see a fundamental purposive difference between the legal forms, was always going to be a choice made upon a number of grounds, including — crucially — its consequences for relations with third parties. One commentator bemoaned that “in many cases, the choice of the SA form (rather than of the SARL form) is motivated by social, financial and even psychological considerations, rather than the original justification for this corporate form” (Guennouni 2004).11 But that, of course, is precisely the point: to truly understand the incentives facing firms in choosing their legal status, it is necessary to model the effect that such choice would have upon others, particularly upon banks.
11
Original quote: “Dans de nombreuses hypoth`eses, en effet, le choix de la soci´et´e anonyme (plutˆot que de la SARL) est motiv´e par des consid´erations d’ordre social, fiscal, voire psychologique et non par la raison d’ˆetre originelle de cette forme de soci´et´e.”
3. THEORETICAL MODEL
3.1
Motivating literature: Credit constraints and information
This chapter develops a theoretical model for the role of legal status in bank lending decisions. That model will draw attention to the role of legal status as an informative signal in a context where information between bank and firm is asymmetric. This section first anticipates that analysis by reviewing past literature to show why the issue of credit constraints is properly considered as a problem of asymmetric information. The concept of ‘credit constraints’ has no single agreed meaning in the literature. However, a common concept can be distilled, and is used here: ‘credit constraints’ are understood as describing a situation where a firm with a potentially profitable opportunity cannot obtain sufficient bank credit for that opportunity, either because credit is refused altogether or because an inadequate quantity is provided.
As early as Jaffee and Modigliani (1969), it was recognised that credit constraints can arise in a market of rational agents if that market is characterised by asymmetric information. However, such early research dealt awkwardly with asymmetric information; for example, by invoking institutions encouraging banks “to limit the spread between rates”, without modelling why such behaviour would be rational. Spence (1973) provided sounder methodological foundations, in modelling the role of education as a costly signal or workers’ unobservable characteristics. Spence’s work has been analysed and summarised extensively; for present purposes, we merely note three key insights for present modelling. First, if legal status is to be an informative signal of un-
3. Theoretical model
14
observable firm characteristics, the decision to “invest in the signal” (Spence 1973, 358) must be more costly for poorer quality firms: “a signal will not effectively distinguish one applicant from another, unless the costs of signalling are negatively correlated with productive capacity” (Spence 1973, 358). Second, though Spence focussed upon a continuous signal (education) for a discrete unobservable type (productive capacity), he acknowledged immediately that the signalling framework could equally extend to discrete signals and to continuous unobservable types (367–368). This is particularly important for modelling legal status as a signal; legal status is a binary variable and, to appreciate fully the ‘pooling’ effects imposed by a binary signal, it will be appropriate to model firm quality as a continuous variable.1 Finally, though he did not use the term as such, Spence provided a persuasive justification for the use of what is now termed “Perfect Bayesian Equilibrium” (‘PBE’) as the relevant solution concept; the concept will be outlined and justified further shortly.
Jaffee and Russell (1976) was the first direct application of the asymmetric information framework to the issue of credit constraints. The authors used a two-period consumption model with different classes of borrowers — indistinguishable a priori — to formalise the notion that banks face a variety of different loan prospects, so that lending decisions involve inferring imperfectly the quality of the borrower. This work foreshadowed the landmark exposition of credit rationing as a phenomenon of imperfect information: Stiglitz and Weiss (1981). The primary contribution of Stiglitz and Weiss (1981) was to consider the role of interest rates and of collateral as screening devices. Though we do not model those terms here, the paper nonetheless suggests a number of insights. Fundamentally, the authors argued that the distribution of bor1
Indeed, there is a literature specifically upon the use of such ordinal signals of cardinal information: the literature on ‘coarse information’. For example, Meyer (1991) considers the strategic use of such information in repeated trials to explain the use of biased contests by firm management. Were we modelling a repeated game with agent memory, this literature would suggest a raft of additional considerations, relating to banks strategically inducing repeated changes in legal status to reveal hidden information. However, given that our framework is static, the issue need not further concern us here.
3. Theoretical model
15
rowers varies as the interest rate changes, because “higher interest rates induce firms to undertake projects with lower probabilities of success but higher payoffs when successful” (393). Though Stiglitz and Weiss did not cite his work, Adam Smith had recognised the same point two centuries earlier in The Wealth of Nations: Smith had worried that, if interest rates were too high, the distribution of borrowers would primarily comprise “prodigals and projectors, who alone would be willing to give this high interest” (Smith 1776, 356-357).2 This insight is fundamental to the current research, albeit by analogy: if legal status is relevant to determining loan outcomes (for example, because a bank decides to lend to ‘high status’ firms but not to those of ‘low status’), then any change to firms’ cost of acquiring ‘high status’ (for example, the introduction of a more stringent legal regime) will change the number and distribution of firms receiving a loan.
Stiglitz and Weiss (1981) prompted a flurry of extensions in the literature, particularly concerning the role of collateral as a signal of borrower wealth: see Wette (1983), Bester (1985), Bester (1987) and Besanko and Thakor (1987). These additional results are not directly useful to the current research (which does not deal directly with issues of wealth or collateral); rather, the important point is that, despite arguments to the contrary, the results and methodology of Stiglitz and Weiss (1981) withstood the consequent critique.
3.2
Conceptual framework
It is a mark of the significance of Stiglitz and Weiss (1981) that no further transformational contribution to the theory of credit constraints under imperfect information has arisen since. (There has, however, been significant research upon the empirical estimation of credit constraints; this will be considered in Chapter Four.) However, the 2
In referring to this quote, Sen clarifies that “[t]he term ‘projector’ is used by Smith not in the neutral sense of ‘one who forms a project,’ but in the old pejorative sense”; see Sen (1999, 324).
3. Theoretical model
16
literature surveyed suffices to identify the key methodology necessary to model the role of legal status in determining credit rationing. Following Jaffee and Russell (1976), we shall explain credit constraints as arising from banks using imperfect information about borrower quality; Spence (1973) provides the game theoretic context (and the solution concept) for doing that, and for using legal status as a binary signal of a continuous unobservable quality. Taking Stiglitz and Weiss’s (1981) insight, the model shall consider the way in which the distribution of borrowers shifts as the relative cost of signalling changes.
The literature also suggests a number of potential extensions apposite to the current problem. To the author’s knowledge, no model has considered credit constraints as arising from a continuous variable signalled by a discrete variable. However, this seems a natural extension, and one of particular relevance for any question involving a legal or administrative demarcation; unlike quantities like education (Spence 1973), or interest rate or collateral (Stiglitz and Weiss 1981), many informative signals (like the SA/SARL distinction) are required by the institutional context to be discrete. This does not pose any significant modelling challenge; however, it is argued that a number of interesting and informative insights emerge. These will be highlighted as the model is developed.
3.3
Modelling a single, unobserved, continuous type 3.3.1
Developing the model
Players: Following the general approach of the theory of contracts, the model seeks a simple partial-equilibrium description of representative agents (Salani´e 1997, 2), concerned with individually rational behaviour (ignoring coalitional actions) (Spence 1973, 364). Thus, the players are Firm and Bank; further, to incorporate a role for uncertainty, we allow a move by Nature.
3. Theoretical model
Game structure and strategies:
17
The interaction is modelled as a signalling game,
with the informed party (Firm) moving before the uninformed party (Bank). The alternative, of course, would be a screening game, with Bank moving first; the choice of the signalling structure reflects an intuitive sense that, as between a firm’s choice of legal status and the bank’s choice of loan conditions, it is the firm’s choice that is made in anticipation of the bank’s, rather than the converse.
The game structure and strategies are therefore as follows. 1. Nature chooses a univariate ‘quality’ for Firm, Q: Q ∼ FQ∈[Q,Q] (q).3 For simplicity, the cumulative density function FQ (·) (and thus the probability density function fQ (·)) is common knowledge among all players.4 2. Firm observes Q and chooses a status, φ ∈ {H, L}. The status φ = H represents the ‘High’ status, SA form; φ = L represents the ‘Low’ SARL form. It will emerge in solving the model that φ may have a signalling role; thus, when we discuss in this context the decision to ‘invest in the signal’, we refer to Firm choosing φ = H rather than φ = L. 3. Bank observes Firm status φ and chooses an action, γ ∈ {Lend, Refuse}. For simplicity, we assume in this model that Bank may offer only one type of loan contract; thus, Bank’s only strategic choice is whether or not to provide that contract.5 3
By this notationR — used throughout — it is meant that FQ (·) is the cumulative density function for Q: q Pr(Q ≤ q) = Q fQ (ϕ) dϕ, where fQ (·) is the probability density function for Q.
4
That is, “not only is it mutual knowledge but also each individual knows that all other individuals know it, each individual knows that all other individuals know that all the individuals know it, and so on” (Osborne and Rubinstein 1994, 73). Common knowledge is assumed to ensure that the solution of the model is driven solely by the signalling effect of legal status (rather than, for example, by complicating considerations of the players’ inferences about each other’s knowledge). This assumption is less restrictive than it first appears. When, in the bivariate case, we introduce an observable component to the quality of Firm, there is nothing to prevent Bank varying the terms of its loan contract systematically with that observable component (for example, offering larger loans to larger firms). What is ruled out here for simplicity is that Bank offers one type of loan where Firm signals φ = H and another type where φ = L; Bank must either offer the same contract for both φ = H and φ = L or Lend for one choice of φ (for example, φ = H) but Refuse for the other (for example, φ = L).
5
3. Theoretical model
18
Payoffs: Payoffs for Firm of type Q accrue through three conceptually distinct sources: • Firm’s profit from production, determined by Firm’s quality (Q): Y (Q). • Firm’s additional benefit from receiving a loan: L(Q). This is assumed to be an increasing function of firm quality: L(Q) > 0 (with L0 (Q) > 0). This reflects an intuition that firms of higher quality can benefit more from a given loan. • Firm’s relative cost of choosing φ = H over φ = L: C(Q). This is the relative cost to Firm of ‘investing in the signal’; it is assumed to be a positive but decreasing function of firm quality: C(Q) > 0 (with C 0 (Q) < 0). This reflects the intuition that choosing SA over SARL form is costly, but more costly for firms of lesser quality. Together, restrictions on L0 (Q) and C 0 (Q) are necessary to capture the intuition that, as the quality of Firm increases, it is relatively more profitable to choose φ = H for the purpose of accessing credit; this reflects Spence’s (1973) point that the cost of a signal must be negatively correlated with unobserved quality in order that the signal be informative.
Consider, in contrast, payoffs to Bank. These are modelled as accruing from two conceptually distinct sources: • Bank’s profit from lending to Firm of type Q: P (Q). This is assumed to be an increasing function of firm quality (such that Bank benefits more by lending to Firm of higher quality): P 0 (Q) > 0. • Bank’s additional ‘transparency gain’ from lending to Firm of status φ = H: T (Q). This captures the notion that, quite apart from signalling effects, Bank may find it more profitable to lend to ‘high’-type firms because of the higher
3. Theoretical model
19
requirements of transparency (and governance, etc). That is, it captures the intuition that an agent that invests in a signal may confer a benefit upon a principal independent of the signalling effect; the analogy in the ‘education signalling’ literature is that, by investing in education, an employee may become more productive for his or her employer separately from signalling his or her aptitude for work or education. Figure 3.1 illustrates the structure and payoffs.6 The payoff functions are understood to operate as von Neumann-Morgenstern utility functions; this is completely standard and allows naturally for an expected utility framework.
Fig. 3.1: The structure of the game 6
Note that, for completeness, a payoff of zero has arbitrarily been assigned to Nature.
3. Theoretical model
3.3.2
20
Solving the model
The solution concept relevant to this game is Perfect Bayesian Equilibrium (‘PBE’). This requires (in the context of this specific model)7 that: 1. Firm’s choice of φ maximises its payoff given Q, γ and Bank’s beliefs; 2. Bank’s choice of γ maximises its expected payoff given φ and Bank’s beliefs; 3. At information sets reached with positive probability when the strategy is played, beliefs are formed according to the strategy and Bayes’ Rule; 4. At information sets reached with probability zero when the strategy is played, beliefs may be arbitrary, but must use Bayes’ Rule to the extent possible.8 Perfect Bayesian Equilibrium is the natural solution concept for a signalling game of this form. Spence (1973, 360) provided the fundamental justification: “To avoid studying a system in a continual state of flux, it is useful to look for nontransitory configuration [sic] of the feedback system. The system will be stationary if the [principal] starts out with conditional probabilistic beliefs that after one round are not disconfirmed by the incoming data they generated. . . An equilibrium is a set of components in the cycle that regenerate themselves.”
The solution to this model does not transpire to be particularly difficult, nor intuitively perplexing. However, it benefits from a somewhat circuitous method of solving; namely, to solve first under the assumption of a separating equilibrium, to derive from that result the legitimate restriction on Bank beliefs, then to solve for potential pooling equilibria. This approach is without loss of generality. First, clarify the meaning of pooling and separating equilibria in this context. 7
8
The concept is applied here only in the context of the particular model; our concern here is with the application of the solution concept to the credit constraint problem. Thus, we do not pursue here a general formalisation of the concept — for example, see Osborne and Rubinstein (1994, 232-233). The notion of beliefs in this framework will be developed further shortly.
3. Theoretical model
21
Definition 1: Pooling and separating equilibria in the univariate continuous case: • If there is an equilibrium in which Firm chooses the same φ irrespective of Q, it is a pooling equilibrium; • If there is an equilibrium in which Firm conditions its choice of φ upon Q, it is a separating equilibrium.
Solving for separating equilibria In this section, we solve for the restrictions necessary to support a separating equilibrium. The section proceeds by a number of Propositions; for each, the proof is provided in the appendix to this chapter (pages 49–50).
Firm’s best response under separating equilibrium: We assumed earlier that L(Q) > 0 and C(Q) > 0. Thus, assuming a separating equilibrium, Firm has a unique optimal strategy. First, consider the way in which Bank must be acting in order for there to be a separating equilibrium. Proposition 1: For any separating equilibrium, Bank chooses Lend for φ = H and Refuse for φ = L. Given Bank’s strategy, it is straightforward to understand Firm’s best response. Proposition 2: For any separating equilibrium, Firm has a unique best response to Nature: choose φ = H if and only if L(Q) ≥ C(Q), else choose φ = L.9 9
It is immediately obvious that this strategy is not, literally, a unique best response: an equally good response would be to reverse the strict and weak inequality, ie choose φ = H if and only if L(Q) > C(Q), else choose φ = L. However, in the continuous case, equality occurs with probability zero; thus, it is without loss of generality that we impose that, where Firm is indifferent, it will choose φ = H and where Bank is indifferent, it will choose Lend. In this case and every subsequent case, discussion of ‘unique’ best responses must be construed accordingly, ie unique but for the possibility of reversing the strict and weak inequalities.
3. Theoretical model
22
Thus, if there is a separating equilibrium, Firm has a unique best response to Nature for all Q. We now claim that this best response is to follow a ‘cut-off’ strategy.
Definition 2: The cut-off value, Q∗ • Define Q∗ such that L(Q∗ ) = C(Q∗ ). At this stage, assume the existence of Q∗ ∈ (Q, Q). Then we may be more specific about Firm’s optimal strategy. Proposition 3: If there is a separating equilibrium, Firm’s unique best-response to Nature is to follow a ‘cut-off strategy’: choose φ = H ∀ Q ≥ Q∗ ; choose φ = L otherwise. This result is a close analogue to Theorem 1 in Stiglitz and Weiss (1981, 396), albeit arising from a different modelling approach. In their model, Stiglitz and Weiss showed that a firm’s optimal decision to borrow follows a cut-off strategy based upon firm quality; in a separating equilibrium, Proposition 3 here amounts to precisely the same result.
Summary: Firm’s best response to Nature is straightforward: • If there is a separating equilibrium, choose φ = H ∀ Q ≥ Q∗ , else choose φ = L.
To close the solution for the separating equilibrium case, we need to consider the form that Bank’s payoffs must take to induce Bank to indeed optimally choose Lend for φ = H and Refuse for φ = L; that is, we need to solve for restriction’s upon Bank’s ‘incentive compatibility’ condition.
3. Theoretical model
23
Bank’s incentive compatibility condition for a separating equilibrium: It has been shown that, under a separating equilibrium, Bank will Lend whenever φ = H is observed and Refuse whenever φ = L is observed. For Bank to behave optimally in this way, it must be that:10
πBank (γ = Lend, φ = H) ≥ πBank (γ = Refuse, φ = H) πBank (γ = Lend, φ = L) < πBank (γ = Refuse, φ = L).
Bank’s unconditional payoffs are simply its expectations of conditional payoffs (outlined earlier), across the range of Q under consideration. Given that Firm has a unique best response to adopt a cut-off strategy, Bank correctly infers — in the separating equilibrium case — that φ = H implies Q ≥ Q∗ while φ = L implies Q < Q∗ . To understand why, we must consider the application of Bayes’ Rule. If there exists a separating equilibrium, Perfect Bayesian Equilibrium requires that Bank form its beliefs consistently with that equilibrium. Thus, following the preceding reasoning, Bank must believe, in the case of a separating equilibrium, that Q ∼ FQ|Q∈[Q∗ ,Q] (Q∗ , Q) if φ = H is observed, and that Q ∼ FQ|Q∈[Q,Q∗ ) (Q, Q∗ ) if φ = L is observed, where the notation fQ|Q∈[a,b] (·) refers to the probability distribution of Q conditional upon the restriction that Q ∈ [a, b], ie using Bayes’ Rule, fQ|Q∈[a,b] (q) =
fQ (q) FQ (b)−FQ (a) .
That is, if there is a separating equilibrium, Bank forms its belief of the distribution of Q by using the observed signal to condition upon whether Q ≥ Q∗ or Q < Q∗ . Thus, Bank’s payoffs are as follows. 10
Recall that we impose that Bank will Lend where it is indifferent; thus, the second inequality is strict.
3. Theoretical model
Z
24
Q
f (q) · (P (q) + T (q)) dq Q∗
πBank (γ = Lend, φ = H) =
1 − FQ (Q∗ ) Z
∗
Q
f (q) · P (q) dq πBank (γ = Lend, φ = L) =
Q
FQ (Q∗ )
πBank (γ = Refuse, φ = H) = 0 πBank (γ = Lend, φ = H) = 0
Therefore, Bank’s incentive compatibility conditions to support a separating equilibrium are: Z
Q
f (q) · (P (q) + T (q)) dq ≥ 0
(3.1)
Q∗
Z
Q∗
f (q) · P (q) dq < 0
(3.2)
Q
These incentive compatibility constraints have intuitive appeal: in order for Bank to Lend to φ = H but not to φ = L, it must be profitable for it to Lend to Firm of type Q ≥ Q∗ (including the additional ‘transparency payoff’, given that these firms invest in the signal), but not profitable to Lend to Firm of type Q < Q∗ (not including the ‘transparency payoff’, because those firms do not invest in the signal). The necessary conditions for the existence of a separating equilibrium are therefore that there exists some Q∗ ∈ (Q, Q] such that L(Q) > C(Q) ∀ Q > Q∗ , and that these two incentive compatibility conditions are met.
3. Theoretical model
25
Solving for pooling equilibria Consider, then, the case where all types of Firm choose the same φ; that is, where either all invest in the signal or none do so. In this section, we consider the strategies and parameter values to support such an equilibrium. Specifically, the section imposes a restriction upon Bank’s beliefs away from the equilibrium path. Consequent upon this restriction, the section then solves for various forms of pooling equilibrium. The potential equilibria are outlined and justified here; they are summarised and synthesised shortly.
Bank beliefs in pooling equilibria: The application of Bayes’ Rule in the separating equilibrium case was just considered. But what of a pooling equilibrium? Clearly, Bayes’ Rule does not directly assist in that case; if every type of Firm optimally chooses the same φ, observing that φ is not informative to Bank of Q. However, what if Firm strays from the equilibrium choice of φ? Posit a pooling equilibrium on φ = Φ. Then, if φ 6= Φ, the game is off the equilibrium path. In this context, Perfect Bayesian Equilibrium allows Bank’s beliefs to be formed completely arbitrarily in this situation. However, it is both useful and intuitive in this game if we restrict Bank’s beliefs in pooling equilibrium from the outset. The restriction used is the ‘intuitive criterion’ of Cho and Kreps (1987). This is “essentially a stability requirement about out-of-equilibrium beliefs: it says that when a deviation is dominated for one type of player but not the other one, this deviation should not be attributed to the player for which it is dominated” (Bolton and Dewatripont 2005, 107–108). In this context, that criterion means that, if only Firm of type Q ∈ X would, in a separating equilibrium, have incentive to choose φ = Φ, and φ = Φ is observed in a pooling equilibrium on φ 6= Φ, Bank must believe that Q ∼ FQ|Q∈X (q). That is, the intuitive criterion imposes here that, if deviation is observed in a pooling equilibrium, Bank believes that the deviating Firm must be of a type for which deviation would be profitable in a
3. Theoretical model
26
separating equilibrium.
It was shown earlier that, in a separating equilibrium, Firm has an incentive to choose φ = H only if Q ≥ Q∗ . Thus, Firm with Q < Q∗ never has an incentive to choose φ = H, whereas Firm with Q ≥ Q∗ has that incentive if it wrongly anticipates a separating equilibrium.11 Thus, in a candidate pooling equilibrium on φ = L, the intuitive criterion imposes that Bank believe that, if φ = H is observed,
Q ∼ FQ|Q∈[Q∗ ,Q] (q).
(3.3)
In a candidate pooling equilibrium on φ = H, the observation that φ = L allows no further restriction on beliefs since, as will be shown, a pooling equilibrium on φ = H implies that it is never profitable for Firm of any type to deviate. We proceed, then, to consider various candidate pooling equilibria.
Pooling on φ = H: Posit now a pooling equilibrium on φ = H. That is, consider an equilibrium in which all types of Firm choose φ = H.
Consider first the strategy that Bank must adopt to support such an equilibrium. Proposition 4: For any pooling equilibrium on φ = H, Bank chooses Lend for φ = H and Refuse for φ = L. Given that strategy, we can understand the parameter restrictions necessary to ensure that Firm always chooses φ = H. 11
The notion of ‘belief’ is used imprecisely here. Clearly, we refer to Firm wrongly understanding the nature of the equilibrium path. This clearly does not refer to a ‘belief’ in the sense of an assigned probability, ie the sense in which the notion is generally used in Perfect Bayesian Equilibrium, including elsewhere in this note.
3. Theoretical model
27
Proposition 5: If there is a pooling equilibrium on φ = H, L(Q) − C(Q) ≥ 0 ∀ Q ∈ [Q, Q]; that is, there exists no Q∗ ∈ [Q, Q] such that L(Q∗ ) = C(Q∗ ). Thus, such an equilibrium could be supported if every type of Firm found it profitable to invest in the signal to receive a loan. In order that this actually induce the choice of φ = H, it must be that Bank would rationally Refuse if φ = L were observed (as outlined in Proposition 4). Consider, then, Bank’s incentive compatibility requirement.
Bank’s incentive compatibility requirement for pooling on φ = H: We know that, in order for there to be a pooling equilibrium on φ = H, Bank must Lend for φ = H and Refuse for φ = L. For this to be compatible with Bank’s incentives, it must be that:
πBank (γ = Lend, φ = H) ≥ πBank (γ = Refuse, φ = H)
(3.4)
πBank (γ = Lend, φ = L) < πBank (γ = Refuse, φ = L).
(3.5)
It was also shown earlier that there must exist no Q∗ ∈ [Q, Q] such that L(Q∗ ) = C(Q∗ ). Thus, Bank’s incentive compatibility condition to support a pooling equilibrium is considered over the entire support Q ∈ [Q, Q]: Z
Q
fQ (q) · (P (q) + T (q)) dq ≥ 0 Q
Z
Q
fQ (q) · P (q) dq < 0. Q
That is, pooling on φ = H can only be supported if every type of Firm can ‘afford’ to invest in the signal, and if it is profitable for Bank to Lend to Firm signalling φ = H only because of the ‘transparency effect’. This makes intuitive sense; after all, for a pooling equilibrium on φ = H, φ is not an informative signal, precisely because every
3. Theoretical model
28
type of Firm must be able to afford it. Thus, the choice of φ = H must be justified, if at all, by the non-signalling benefits that the choice confers upon Bank.
We turn, then, to consider the necessary restrictions for a pooling equilibrium upon φ = L; that is, an equilibrium in which all types of Firm choose φ = L. In the case of pooling upon φ = H, it was just shown that there must be no Q∗ ∈ [Q, Q]. However, it transpires that, in the case of pooling on φ = L, three scenarios are possible: the scenario where there is some Q∗ ∈ [Q, Q] (that is, some types of Firm would choose φ = H if Bank would consequently Lend), and two scenarios where there is no Q∗ ∈ [Q, Q] (scenarios where no type of Firm ever finds it profitable to choose φ = H, and where all types do). We consider each scenario in turn. Pooling on φ = L with Q∗ ∈ (Q, Q]: Now posit a pooling equilibrium on φ = L with the existence of some Q∗ ∈ (Q, Q], ie so that L(Q) > C(Q) ∀ Q > Q∗ . Proposition 6: For any pooling equilibrium on φ = L with the existence of some Q∗ ∈ (Q, Q], Bank will Refuse for φ = H. Bank may Lend or Refuse for φ = L. This is intuitively straightforward: if some types of Firm would choose φ = H in order to induce Bank to Lend, a pooling equilibrium on φ = L can only be supported if Bank will optimally Refuse for φ = H. To understand what Bank will optimally do in a pooling equilibrium on φ = L for observed φ = H, we must consider Bank’s off-equilibrium-path beliefs.
Proposition 7: If φ = H is observed, Bank will believe that Q ∼ FQ|Q∈[Q∗ ,Q] (q). Given this, it is straightforward to identify Bank’s incentive compatibility requirement.
3. Theoretical model
29
Bank’s incentive compatibility requirement for pooling on φ = L with Q∗ ∈ (Q, Q]. It was shown in Proposition 6 that Bank must Refuse for φ = H. For this to be compatible with Bank’s incentives it must be that:
πBank (γ = Lend, φ = H) < πBank (γ = Refuse, φ = H).
Thus, given the support [Q∗ , Q] set out in Proposition 7, Bank’s incentive compatiZ Q fQ (q) · (P (q) + T (q)) dq < 0. bility condition to support a pooling equilibrium is Q∗
That is, Bank must not anticipate profitably choosing to Lend if Firm deviates to φ = H.
Pooling on φ = L with L(Q) > C(Q) ∀ Q ∈ [Q, Q]: Posit a pooling equilibrium on φ = L with L(Q) > C(Q) ∀ Q ∈ [Q, Q]. That is, consider the situation in which it is profitable for all types of Firm to choose φ = H, if Bank were to Lend contingent upon observing φ = H. It is straightforward, then, that to support a pooling equilibrium upon φ = L, Bank must not Lend upon observing φ = H. Proposition 8: For any pooling equilibrium on φ = L for which L(Q) > C(Q) ∀ Q ∈ [Q, Q], Bank will Refuse if φ = H is observed. In that case, it is similarly straightforward that, if some type of Firm (wrongly) chooses φ = H, the choice will be uninformative of Q; this allows Bank’s incentive compatibility condition to be derived. Proposition 9: If φ = H is observed, Bank will believe that Q ∼ FQ (Q, Q). Bank’s incentive compatibility requirement for pooling on φ = L with L(Q) > C(Q) ∀ Q ∈ [Q, Q]: It was shown in Proposition 8 that Bank must Refuse for φ = H. For this to be compatible with Bank’s incentives it must be
3. Theoretical model
30
that πBank (γ = Lend, φ = H) < πBank (γ = Refuse, φ = H). Given the support [Q, Q] in Proposition 9, Bank’s incentive compatibility condition to support a poolZ Q ing equilibrium is therefore fQ (q) · (P (q) + T (q)) dq < 0. Thus, if every type Q
of Firm will profitably choose φ = H if Bank will consequently Lend, a pooling equilibrium upon φ = L is supported by Bank anticipating a loss in that case.
Pooling on φ = L with L(Q) < C(Q) ∀ Q ∈ [Q, Q]: Posit a pooling equilibrium on φ = L with L(Q) < C(Q) ∀ Q ∈ [Q, Q]. That is, consider the case in which it is never profitable for any type of Firm to choose φ = H, even if Bank will consequently Lend. Proposition 10: If L(Q) < C(Q) ∀ Q ∈ [Q, Q], there exists no Q such that Firm of type Q will deviate to φ = H.
Bank’s incentive compatibility requirement for pooling on φ = L with L(Q) < C(Q) ∀ Q ∈ [Q, Q]: Proposition 10 implies that there is no binding incentive compatibility requirement for this case, because Proposition 10 holds irrespective of Bank’s response. Nonetheless, it bears noting the basis for Bank’s lending deciZ Q sion: Lend if and only if πBank (γ = Lend, φ = L) ≥ 0 ⇔ f (q) · P (q) dq ≥ 0. Q
Summary of equilibria The possibilities reduce to three distinct Cases, which together are mutually exclusive and collectively exhaustive.
Case 1: Where it is never profitable for Firm to be φ = H.
3. Theoretical model
31
This case implies that L(Q) < C(Q) ∀ Q ∈ [Q, Q]. In this case, an equilibrium will exist: • Bank will Lend if and only if Z
Q
f (q) · P (q) dq ≥ 0
πBank (γ = Lend, φ = L) = Q
• Bank will Refuse if Z
Q
f (q) · P (q) dq < 0.
πBank (γ = Lend, φ = L) = Q
Thus, the conditions for Bank to Lend and to Refuse are mutually exclusive and collectively exhaustive for Case 1. In both cases, there is a pooling equilibrium on φ = L.
Case 2: Where it is always profitable for Firm to be φ = H, if Bank will Lend to φ = H but not to φ = L.
This case applies if L(Q) ≥ C(Q) ∀ Q ∈ [Q, Q]. As explained earlier (see Equations 3.4 and 3.5), this pooling equilibrium on φ = H can only be supported if Bank will Lend for φ = H but Refuse for φ = L.
Thus, as shown previously, the incentive compatibility requirements for this outcome are as follows.
3. Theoretical model
32
• Bank will Lend if and only if: Z
Q
f (q) · (P (q) + T (q)) dq ≥ 0
πBank (γ = Lend, φ = H) = Q
Z
Q
f (q) · P (q) dq < 0
πBank (γ = Lend, φ = L) = Q
This will support a pooling equilibrium on φ = H. • Otherwise, no pooling equilibrium on φ = H will exist. If no pooling equilibrium on φ = H will exist, Firm will choose φ = L ∀ Q, since there is no Q∗ ∈ [Q, Q] that could otherwise be used to support a separating equilibrium. Thus, as in Case 1, – Bank will Lend if and only if Z
Q
f (q) · P (q) dq ≥ 0
πBank (γ = Lend, φ = L) = Q
– Bank will Refuse if Z
Q
f (q) · P (q) dq < 0.
πBank (γ = Lend, φ = L) = Q
Thus, in either of these later cases, there will be a pooling equilibrium on φ = L. These outcomes, then, are mutually exclusive and collectively exhaustive for Case 2.
Case 3: Where it is profitable for Firm of type Q ≥ Q∗ to be φ = H, if Bank will Lend to φ = H but not to φ = L, but it is not profitable for Firm of type Q < Q∗ .
This case applies if L(Q) ≥ C(Q) ∀ Q ∈ [Q∗ , Q] and L(Q) < C(Q) ∀ Q ∈ [Q, Q∗ ).
3. Theoretical model
33
As shown earlier (see Equations 3.1 and 3.2), the necessary conditions upon Bank’s payoffs for this equilibrium are: • Bank will Lend for φ = H and Refuse for φ = L if and only if: Z
Q
f (q) · (P (q) + T (q)) dq ≥ 0 Q∗
Z
Q∗
f (q) · P (q) dq < 0 Q
This is clearly a separating equilibrium. • Otherwise, no separating equilibrium will exist. If no separating equilibrium will exist, Firm will choose φ = L ∀ Q. (That is, we can rule out immediately the possibility here of a pooling equilibrium on φ = H, since for Firm of type Q < Q∗ , deviation to φ = L is profitable. Since there is neither a separating equilibrium nor a pooling equilibrium on φ = H, Firm of type Q ≥ Q∗ has no incentive to choose φ = H; it chooses φ = L.) Thus, as in Case 1, – Bank will Lend if and only if Z
Q
f (q) · P (q) dq ≥ 0
πBank (γ = Lend, φ = L) = Q
– Bank will Refuse if Z
Q
f (q) · P (q) dq < 0.
πBank (γ = Lend, φ = L) = Q
Thus, in these later cases, there will again be a pooling equilibrium on φ = L. Again, the outcomes specified are mutually exclusive and collectively exhaustive.
3. Theoretical model
34
Summary: This section has presented three Cases. The three Cases themselves are mutually exclusive and collectively exhaustive of the possible forms of L(Q) ≷ C(Q) ∀ Q ∈ [Q, Q], given the restrictions upon functional form set out initially. It has been shown that, in each Case, the model produces mutually exclusive and collectively exhaustive equilibrium outcomes. Thus, for any functional forms L(Q), C(Q), P (Q), T (Q) and FQ (q) meeting the requirements set out, there exists a unique equilibrium outcome, as specified.
3.3.3
Insights from the univariate model
This concludes consideration of the univariate model. The model is informative of the relationship between firm status and access to credit, but not directly useful. There are two closely related reasons for this: (i) its predictions are too sharp: there is no possibility of firms that choose the same status receiving different treatment from a bank; and (ii) there is no role for a commonly observed component to firm quality. A bivariate extension, incorporating both ‘observed quality’ and ‘unobserved quality’, deals with both of these drawbacks. In turning to a bivariate model, we shall rely upon the results and intuition outlined in the univariate model to underpin further analysis; thus, the development of the univariate model — while not directly useful for motivating empirical analysis — will prove essential in underpinning the bivariate model.
3.4
The bivariate case
Motivation for the bivariate model:
The bivariate model seeks to formalise the in-
tuition that, when assessing the quality of a firm, a bank will have access to some information about a firm, in addition to observing its legal status. The combination of
3. Theoretical model
35
this information and the legal status then form the basis of inferring other, unobserved information (eg management quality, general trustworthiness, etc). That is, this extension formalises in this context Spence’s (1973, 369) insight that an uninformed player in a screening game can often profitably condition its action upon ‘observable, unalterable’ characteristics of the informed agent, in addition to acting upon the signal itself.
Of course, were we seeking a more literal representation of the situation, we could develop a K-variable model, with each of the K variables representing a component of Firm quality, with some variables being observable and others unobservable (for example, size, number of employees, firm age, production sector, management quality, etc). However, this would be unnecessarily complicated. We develop from the univariate to the bivariate case to allow a distinction between observed and unobserved information; a distinction that will prove fundamental for motivating empirical analysis. The bivariate case, therefore, is the simplest case capturing the key relationship that we seek to explore.
3.4.1
Developing the bivariate model
Distinguishing observed and unobserved variables The basic structure and payoffs remain the same. However, we now replace the univariate Q with the bivariate double Q = {Qo , Qu }, where Qo refers to an ‘observed quality’ index (for example, representing the effect of variables such as age of firm, number of employees, etc), and where Qu refers to an ‘unobserved quality’ index (for example, capturing notions of management quality, trustworthiness, accuracy of market perception, etc). Now, assume that Nature chooses Q according to a bivariate distribution Q ∼ FQ∈{(Qo ,Qo ),(Qu ,Qu )} (q), where q = {qo , qu }. That is, making explicit the Rq Rq form of the cumulative distribution FQo ,Qu (qo , qu ) = Qo Qu fQo ,Qu (s, t) dt ds. o
u
3. Theoretical model
3.4.2
36
Solving the model
Given the results derived in the univariate case, the bivariate case may be solved very briefly — for in the bivariate case, all of univariate results hold as a matter of direct application, conditional upon a given observed quality, Qo . (That is, for any given Qo , the game is precisely equivalent to the univariate game, with Qu taking the place of the univariate Q.) Thus, extending the univariate case to the bivariate case really requires only that we understand how Bank may use the observed Qo to draw inference about the unobserved Qu . In doing so, it bears clarifying the meaning of separating and pooling equilibria as they are used in the bivariate case. Definition 3: Pooling and separating equilibria in the bivariate continuous case • If, for a given Qo , there is an equilibrium in which Firm chooses the same φ irrespective of Qu , it is a pooling equilibrium; • If, for a given Qo , there is an equilibrium in which Firm conditions its choice of φ upon Qu , it is a separating equilibrium. That is, in the bivariate case, we consider whether there is a pooling or separating equilibrium conditional upon observed Qo . In effect — as the subsequent examples will reveal — we consider pooling and separating ‘regions’ across the range of Qo values. Further, for a separating equilibrium (that is, for ‘regions’ of Qo supporting separating equilibria), we can consider the analogy to the cut-off value Q∗ in the bivariate context. Definition 4: The cut-off value in the bivariate case, Q∗u (Qo ) • Just as we previously defined Q∗ such that L(Q∗ ) = C(Q∗ ), let us now define Q∗u (Qo ) such that L(Qo , Q∗u (Qo )) = C(Qo , Q∗u (Qo )). That is, Q∗u defines the ‘cut-off’ value of Qu such that, for a given Qo and Qu ≥ Q∗u , Firm will optimally choose φ = H in a separating equilibrium. (It is not necessary to
3. Theoretical model
37
prove such a cut-off result again; conditional upon any particular Qo , the result holds as a trivial application of Proposition 3.)
With these definitions in place, the solution to the bivariate case is merely an application of the univariate solution: one of the three Cases set out in solving the univariate model will hold, depending upon the particular parameter values, in precisely the same way that they did under the univariate specification. All that remains is to allow Bank to use optimally the value of Qo to draw inference about the distribution of Qu . This requires that, in the bivariate case, the probability density function for Q is replaced by the conditional probability density function for Qu given Qo :
fQu |Qo (qu |qo ) =
fQo ,Qu (qo , qu ) . fQo (qo )
With that extension, the bivariate model is solved: for all of the results in the univariate model (including, crucially, the three mutually exclusive and collectively exhaustive Cases set out at the end of that model) hold in the bivariate case, for any given Qo , with the relevant functions Y (·), L(·), C(·), P (·) and T (·) now operating upon the bivariate double Q = {Qo , Qu } rather than the univariate Q. (In each case, if the univariate function was increasing (decreasing) in Q, the bivariate equivalent is assumed to be increasing (decreasing) in both Qo and Qu separately.)
3.4.3
Applying the bivariate model
This result — that the solution to the univariate case extends naturally to the bivariate case — is important for reassuring us that introducing an observed component Qo does not erode the analytical foundations for the solution. However, to understand intuitively the insights suggested by the bivariate case, it is useful to apply the bivariate model under a particular specifications for the distribution of Q. Before proceeding
3. Theoretical model
38
further, let us impose for simplicity the requirement that Bank be risk neutral, so that E[πBank (Qo , Qu , φ)] ≡ πBank (Qo , E[Qu |Qo , φ], φ). This restriction is necessary to ensure the tractability of the model, by focussing attention upon Bank’s inference about Qu . Clearly, this restriction is not literally true; however, in the context of a model capturing the tension between Bank and Firm, it is a reasonable assumption given that banks are surely more risk-neutral than loan recipients. Further, the assumption is standard for the literature in this area: see Jaffee and Russell (1976, 658).
The simplest informative example is the bivariate normal distribution; this provides a relatively straightforward illustration that allows for covariance between Qu and Qo and for closed-form solutions for E(Qu | Qo , φ). Suppose, then, that {Qu , Qo } is distributed bivariate normal, so that: 2 µ Q u σu u , ∼ N ρσu σo µo Qo
ρσu σo . σo2
Now, by well-known results that we need not pause to rehearse here (closely related to the derivation of the tobit estimator: see Tobin (1958)), we know that: Qu |Qo ∼ N
µu + ρ
σu (Qo − µo ), σu2 (1 − ρ2 ) . σo
Further, for any random variable Y ∼ N (µY , σ 2 ), we know that
E(Y |Y > α) = µY + σ
φ Φ
µY −α σ µY −α , σ
where φ and Φ are the normal pdf and cdf respectively.
3. Theoretical model
39
We can now solve for the expectation of Qu , conditional upon Qo and φ = H:
E(Qu |Qo = qo , φ = H) = E((Qu |Qo )|(Qu |Qo ) > Q∗u ) = E(Qu |Qo ) + σ(Qu |Qo )
= µu + ρ
p σu (Qo − µo ) + σu 1 − ρ2 · σo
φ
E(Qu |Qo )−Q∗ u σ(Qu |Qo )
Φ
E(Qu |Qo )−Q∗ u σ(Qu |Qo )
∗ u µu +ρ σ σo (Qo −µo )−Qu
u µu +ρ σ σo
φ
√
σu
Φ
1−ρ2
(Qo −µo )−Q∗ u
√
σu
,
1−ρ2
since we know that Qu |Qo is normally distributed. By equivalent reasoning, E(Qu |Qo = qo , φ = L) = µu + ρ
p σu (Qo − µo ) − σu 1 − ρ2 · σo
φ
∗ u µu +ρ σ σo (Qo −µo )−Qu
√
σu
1−Φ
1−ρ2
u µu +ρ σ σo
σu
(Qo −µo )−Q∗ u
√
1−ρ2
This can be neatly illustrated by two diagrams. Figure 3.2 shows a simple ‘firm indifference curve’, in (Qo , Qu ) space; for reasons already explained, Firm will — in the case of a separating equilibrium — signal φ = H if its quality lies on or above the curve, and φ = L if its quality lies below the curve. Figure 3.3, then, graphs a generic illustration of the expectations just derived (with a positive covariance between Qo and Qu , as we would necessarily expect). Notice immediately that the two expectations curves are not equidistant; this will be pursued as a basis for econometric testing in Chapter Four.
Equilibria in the bivariate normal case To consider equilibria in the bivariate normal case, we must specify a relationship for P (Q) and T (Q). In particular, we must consider the level set {Qo , Qu } such that P (Qo , Q∗u (Qo )) + T (Qo , Q∗u (Qo )) = 0. Call this the ‘bank indifference curve’, being the unique level set for which Bank is indifferent about whether to Lend or Refuse, conditional upon Qo and φ. A simple illustration of the bank indifference curve is shown in Figure 3.4; the restrictions earlier placed upon P (·) and T (·) imply that it
.
3. Theoretical model
Fig. 3.2: The ‘firm indifference curve’
Fig. 3.3: Drawing inference in the bivariate normal case
40
3. Theoretical model
41
must slope downwards in (Qo , Qu ) space, as shown. In this and all subsequent figures, we assume for simplicity that T (Qo , Qu ) = 0 ∀ {Qo , Qu }; ie we rule out any additional ‘transparency payoff’ from lending to Firm of status φ = H.
Fig. 3.4: The ‘bank indifference curve’
Finding equilibrium in the model, then, amounts to superimposing the bank indifference curve upon the firm indifference curve and the conditional expectation curves. Figure 3.5 shows this. By allowing for both pooling and separating equilibria, Figure 3.5 provides a general illustration of the possible outcomes; from left to right, the figure illustrates pooling, separating and pooling regions of Qo .
However, this depiction is clearly unsuitable as a description of the role of legal status as a signal in a credit market. Intuitively, one would expect, in a credit market, that firms with lesser observable quality face difficulties of credit constraints, whereas firms with higher observed quality are more likely to receive credit (even if, in due course, they produce non-performing loans). However, as Figure 3.6 shows, this is not the present scenario. Figure 3.6, derived directly from Figure 3.5, shows as ‘credit
3. Theoretical model
42
Fig. 3.5: Separating and pooling equilibria when the bank indifference curve is steeper than the firm indifference curve
constrained’ those regions of (Qo , Qu ) space where Firm does not receive a loan, but which would have been profitable loans for Bank. Conversely, it also shows regions of ‘non-performing loans’, where Firm receives a loan but makes a loss. Figure 3.6 suggests two disturbing features. First, we must be concerned by the two ‘pooling’ regions (that is, the two regions on either end): intuitively, it is difficult to accept that any firm would either be so ‘good’ or so ‘bad’ in its observable characteristics that its unobservable characteristics would never justify investing in the signal. Further, even in the intermediate range where the equilibrium is separating, the scenario lacks intuitive appeal: it is firms with lower Qo that produce non-performing loans, but firms with higher Qo that are credit constrained. This, too, does not seem to match common descriptions of credit constraints.
The key lies in the relative steepness of the firm indifference curve and the bank indifference curve. Figures 3.5 and 3.6 depict the situation where the bank indifference curve is steeper than the firm indifference curve. This corresponds to a situation where
3. Theoretical model
43
Fig. 3.6: Separating and pooling equilibria: Credit constraints and non-performing loans
the signal is relatively uninformative of Qu ; that is, the firm indifference curve is relatively flat because Firm finds it relatively cheap, in terms of Qu relative to Qo , to invest in the signal. Thus, the bivariate extension produces an intriguing insight into understanding the informativeness of a signal: it shows that the informativeness of the signal depends upon changes in the cost of the signal in terms of the unobservable dimension, relative to changes in the cost across the observable dimension.
This explains why Figures 3.5 and 3.6 are not appropriate for analysing the credit market: they describe a situation where the signal under consideration is relatively very uninformative. Consider instead, by way of illustrative example, if the signal observed was not firm legal status but attendance by a firm manager at a religious service (eg the manager observably attends a Mosque every Friday). Religious worship may be taken as some indication of unobservable attributes (such as honesty, diligence, trustworthiness, etc), but is surely a very weak indication. In that case, Figures 3.5 and 3.6 seem quite apposite. Pooling equilibria are observed for the smallest and the largest firms, because the signal is so weak as to be completely uninformative in those cases
3. Theoretical model
44
(relative to observable characteristics); thus, managers of large and small firms have no (pecuniary) incentive to attend the religious service. Among the separating cases, it is the smaller firms that are better able to exploit the signalling opportunity (thus, in some cases, producing non-performing loans): given that Bank is conditioning upon religious attendance, such attendance is a relatively small price to pay for a loan.
Therefore, we must conclude that the case illustrated in Figures 3.5 and 3.6, while interesting for completeness of the exposition, is not a realistic representation of the present scenario. Instead, we must confine our attention to the case where the signal is relatively informative of Qu ; this justifies making the firm indifference curve steeper than the bank indifference curve. Figure 3.7 shows the consequence. The restriction rules out the possibility of pooling equilibria, and reveals that — as one would intuitively expect — it is the firms with lower Qo (eg smaller firms) that may face credit constraints, while firms with larger Qo may produce non-performing loans. It is this figure that most clearly captures the key insight from this modelling methodology. Why do credit constraints exist? In this model, they arise because, under imperfect information, firms that could profitably use a loan are not always able to distinguish themselves from firms that could not; the reason that they cannot always do so is that the informativeness of the available signal of unobserved characteristics, conditional upon observed characteristics, does not match precisely the bank’s concern for unobserved characteristics relative to observed characteristics. In short, this model suggests that credit constraints arise because some firms with poorer observable characteristics (eg younger or smaller firms) find it relatively too expensive to invest in an informative signal, even though they could profitably use bank finance. Conversely, the model suggests that non-performing loans can arise because some firms with better observable characteristics (eg older or larger firms) find it relatively ‘too cheap’ to invest in the signal, even though they cannot profitably use bank finance. In the former case, the
3. Theoretical model
45
bank might wish ex post to have loaned; in the latter case, it regrets ex post that it did. In both cases, asymmetric information — and the imperfect ability of the informative signal to overcome that asymmetry — explains the market failure.
Fig. 3.7: A separating equilibrium when the bank indifference curve is flatter than the firm indifference curve
Conceptualising the treatment in this model: Crucially, the model also provides an intuitive way of considering the effects of the ‘treatment’ (ie legal reform). We expect that the legal reform would increase the costs of signalling φ = H for all types of Firm. Thus, we would expect the legal reform to shift the firm indifference curve ‘north-east’, so that firms that previously were indifferent as to investing in the signal now strictly prefer not to invest. This shift — and its consequence — is shown in Figure 3.8: the legal reform causes a set of firms to switch from ‘high-type’ to ‘low-type’ and, consequently, to lose access to bank credit. In effect, the shift induces some firms to reveal previously hidden information that now reveals them not to have been profitable. (Figure 3.8 omits the expectation curves for simplicity; note that these curves will move to accommodate the shift in Q∗u , but this does not change either the outcome
3. Theoretical model
46
or the implication of the legal change.) Of course, there could be any number of variations to this ‘treatment’. Crucially, by extending the model from the univariate to the bivariate case, we allow the possibility for the treatment to have differential effects depending upon Qo — so, for example, we could model the treatment as affecting ‘lesser’ firms more by imposing a steeper gradient upon the firm indifference curve as well as an outward shift.
Fig. 3.8: A shift in the firm indifference curve
This result is a straightforward illustration of the key insight suggested: namely, that legal status can be informative signal of unobservable characteristics, so that a change in its cost has implications for credit market equilibrium. However, it bears noting an important additional assumption that is made when using the PBE solution in a ‘comparative statics’ context such as this: that neither agent has a memory for the previous equilibrium. If agents were to have such a memory, Bank could condition its action upon Firm’s signal before and after the reform; in effect Bank would receive two signals, φBefore and φAfter , from which to draw inference about Qu . This extension is left as an avenue for further research.
3. Theoretical model
3.5
47
Conclusions and testable hypotheses
Several key conclusions emerge from the model just developed; to the author’s knowledge, all represent contributions — albeit modest — to the existing literature. First, the model shows that, if we accept that the relative cost of acquiring SA status rather than SARL status is negatively correlated with firm quality, legal status can act as a signal of unobserved attributes. Second, the model shows a way of considering the informativeness of a signal where quality is partially observed: a signal is more informative of unobserved attributes if it costs relatively more in terms of unobserved quality than in terms of observed quality For this reason, the extent and distribution of credit constraints and non-performing loans depends upon the difference between the relative cost of the signal and the relative importance of observed and unobserved quality to the lender (that is, as shown, it depends upon the relative positions of the firm indifference curve and the bank indifference curve). Third, the model shows that, if there is asymmetric information and the distribution of firm quality is non-monotonic, a change in signalling costs will induce an impact that varies non-monotonically across firm quality. Thus, for a concave probability density function (as one would generally intuitively expect), a uniform change in signalling costs will induce a concave relationship between change in lending outcome and the bank’s initial perception of firm quality. In turn, these conclusions suggest specific testable hypotheses.
Hypothesis 1: Choosing SA rather than SARL status causes banks to be more willing to provide credit. This hypothesis can be considered across a number of dimensions of credit provision: • Hypothesis 1.1: Choosing SA rather than SARL causes banks to be more willing to grant an overdraft facility. • Hypothesis 1.2: Choosing SA rather than SARL causes banks to be more will-
3. Theoretical model
48
ing to allow a larger overdraft limit. • Hypothesis 1.3: Choosing SA rather than SARL causes banks to provide more short-term debt. • Hypothesis 1.4: Choosing SA rather than SARL causes banks to provide more long-term debt.
Hypothesis 2: The value of choosing SA rather than SARL is due, in part at least, to signalling effects. (That is, the ‘transparency effect’ of choosing the SA status does not completely explain the value of choosing that status.)
The rest of the thesis is concerned with testing these claims.
THEORETICAL MODEL: APPENDIX
Proof of Proposition 1: Suppose that Bank were to choose the same action irrespective of φ. Then a firm choosing φ = H could profitably deviate by choosing φ = L, since C(Q) > 0. Suppose, alternatively, that Bank were to choose Lend for φ = L and Refuse for φ = H. Then a firm choosing φ = H could profitably deviate by choosing φ = L, since L(Q) > −C(Q). Thus, if a separating equilibrium exists, Bank chooses Lend for φ = H and Refuse for φ = L.12 Proof of Proposition 2: Following the previous proposition, we know that, for a separating equilibrium, it must be that Bank will Lend for φ = H and Refuse for φ = L. Thus, following the payoff structure outlined earlier, Firm will optimally choose φ = H if and only if πFirm (γ = Lend, φ = H, Q) ≥ πFirm (γ = Refuse, φ = L, Q) ⇔ Y (Q) + L(Q) − C(Q) ≥ Y (Q) ⇔ L(Q) ≥ C(Q). Proof of Proposition 3: This follows trivially from Proposition 2 and the earlier restriction that L(Q) − C(Q) is strictly increasing in Q. From that restriction, it follows that there exists a unique Q∗ , and that L(Q)−C(Q) > 0 ∀ Q > Q∗ , and vice versa.
Proof of Proposition 4: The proof is the same as for Proposition 1. As shown there, Firm of status φ = H could profitably deviate unless Bank chooses Lend for φ = H and Refuse for φ = L.
Proof of Proposition 5: Proposition 4 showed that Bank chooses Refuse for φ = L; thus, if L(Q) < C(Q) for some Q ∈ [Q, Q], Firm of type Q may profitably deviate from φ = H. Thus, there could be no pooling equilibrium on φ = H. 12
This has a direct analogy in the education screening literature, where ‘high’ types must be ‘bribed’ to reveal their type; nothing is to be gained by ‘bribing’ a ‘low’ type. In this model, a ‘bribe’ to type Φ can be considered as Bank choosing Lend for φ = Φ and Refuse for φ 6= Φ.
3. Theoretical model
50
Proof of Proposition 6: If Bank will Lend for φ = H and Refuse for φ = L, Firm of type Q will profitably deviate to φ = H ∀ Q ∈ [Q∗ , Q]. It is sufficient to prevent such deviation that Bank will Refuse for φ = H.
Proof of Proposition 7: This is a direct application of the Intuitive Criterion, as imposed in Equation 3.3.
Proof of Proposition 8: Since L(Q) > C(Q) ∀ Q ∈ [Q, Q], it will be profitable for Firm of any type Q to deviate to φ = H if Bank will Lend for φ = H.
Proof of Proposition 9: This is a direct application of the Intuitive Criterion, as imposed in Equation 3.3.
Proof of Proposition 10: Irrespective of γ, it will never be profitable for Firm to deviate given L(Q) < C(Q), ie, ∀ Q, L(Q) < C(Q) ⇔ πFirm|Q (φ = H, γ, Q) < πFirm|Q (φ = L, γ, Q).
4. TESTING STRATEGY
4.1
Outline of the testing strategy
This chapter explains and justifies the assumptions and methodology used to test the hypotheses of interest; these hypotheses are then tested in Chapter Six. However, before proceeding to justify the testing methodology, it bears outlining the econometric methods that the chapter seeks to justify. In short, the following methods will be used.
Hypothesis 1.1: Choosing SA rather than SARL causes banks to be more willing to grant an overdraft facility.
The appropriate estimator is the fixed-effect logit. We shall use the fixed-effect logit specification to regress overdraft access (a binary variable) upon legal status and a time dummy (allowing a constant). If the coefficient on legal status is positive and significant, we shall reject the no-effect null hypothesis in favour of Hypothesis 1.1.
Hypothesis 1.2: Choosing SA rather than SARL causes banks to be more willing to allow a larger overdraft limit.
The estimator used is the fixed-effect OLS. We shall use this specification to regress the overdraft limit (a continuous variable truncated below at zero) upon legal status and a time dummy (allowing a constant). If the coefficient on legal status is positive and significant, we shall reject the no-effect null hypothesis in favour of Hypothesis 1.2.
4. Testing strategy
52
Hypothesis 1.3: Choosing SA rather than SARL causes banks to provide more shortterm debt.
This hypothesis can be tested using the two-period panel, but can also be considered using a five-period or six-period panel, using balance sheet data for the several years before the FACS and then the ICA survey. For the two-period panel, we use a fixedeffect OLS; for the longer panel, we use the first-difference OLS for both specifications. In both cases, if the coefficient on legal status is positive and significant, we shall reject the no-effect null hypothesis in favour of Hypothesis 1.3.
Hypothesis 1.4: Choosing SA rather than SARL causes banks to provide more longterm debt.
The methodology is identical to that used for Hypothesis 1.3.
Hypothesis 2: The value of choosing SA rather than SARL is due, in part at least, to signalling effects.
This hypothesis is tested using the insight developed in Chapter Three: if signalling effects are significant, there will be a concave relationship between change in bank credit and the bank’s initial assessment of firm quality. This is tested using kernel regression, simulating the perfect-information null hypothesis (explained shortly) to impute a p-value for the observed relationship. If the p-value is sufficiently small, we shall reject the perfect-information null hypothesis in favour of Hypothesis 2.
4. Testing strategy
4.2
53
Conceptualising the legal reform as a ‘treatment’
This section justifies the testing methodology for Hypotheses 1.1, 1.2, 1.3 and 1.4 by conceptualising the problem of evaluating the impact of the legal reform as an impact evaluation.
To estimate the effect of choosing the SA status rather than the SARL status is to estimate ‘an average partial effect for a binary explanatory variable’; the effect is therefore known as an ‘average treatment effect’ (AT E) (Wooldridge 2002, 603). As outlined previously, there are a number of dependent variables of interest. At this stage, without loss of generality, denote the particular outcome of interest for firm i at time t as yit . More specifically, denote the outcome for that firm as yit0 if the firm chooses SARL status (that is, wit = 0) and as yit1 if the firm chooses SA status (that is, wit = 1). Thus, we follow the treatment effect literature in considering yit0 and yit1 as the counterfactual outcomes for the given firm. The average treatment effect, in population terms, is thus the expected difference in outcomes under the SA and SARL statuses, taken across the population:
AT E ≡ E(y1 − y0 ).
(4.1)
Were legal status assigned randomly, AT E could be estimated consistently by a simple difference-in-means estimator (Wooldridge 2002, 605). However, legal status is clearly chosen by firms; a weaker assumption, then, is that firm legal status (wit ) is chosen as a function of observed firm characteristics (xit ) and of some unobservable firm-specific random variable (ait ), independent of (xit , y0it , y1it ). That is, we allow w = g(xit , ait ). This assumption, sometimes known as ‘selection on observables’ or ‘ignorability of treatment’ would allow the consistent estimation of ATE in a crosssection framework. In this context, we motivate the estimation strategy by imposing a
4. Testing strategy
54
linear relationship; this relationship allows the estimation of AT E by standard fixedeffects estimators. Assume, then, the following structure:
yit = γ0 + γ1 dt + γ2 wit + xit β + ηi + it ,
(4.2)
where xit = (x1it , . . . , xkit ) , β = (β1 , . . . , βk )0 , dt is a time dummy (dt = 0 for FACS and dt = 1 for ICA) and other variables are scalars.
In the context of this linear relationship, the assumption of ‘selection on observables’ amounts to a requirement that the variable ait is not correlated with the firm specific shock ηi + it . That is, in the linear context, the assumption of selection on observables reduces to an assumption that wit is not determined endogenously, so that a basic OLS will estimate consistently the impact of legal status: E(wit (ηi + it )) = 0 ⇔ plim(ˆ γ2,OLS ) = γ2 . This assumption is susceptible to a specific restriction on the information available to the bank: if we are willing to accept that bank lending is a linear function only of variables observed in the survey data (so that the shock ηi + it represents genuine randomness or arbitrariness in outcome, rather than the effect of unmeasured characteristics), the assumption will hold. However, this is almost certainly not the way that banks operate; for example, this excludes the possibility for banks to assess a firm by visiting premises to gain an intuitive assessment of the firm’s competence. Therefore, the assumption that E[wit (ηi + it )] = 0 ought to be relaxed, to allow the possibility that firm status is correlated with unobserved outcome-relevant factors. For example, we must allow the possibility that lending outcomes and firm status both correlate with, say, the director’s fluency in speaking French. Indeed, interviews with Moroccan bank managers for this research strongly suggest that banks base lending decisions, in part at least, upon their assessment of how well loan applicants appear to have considered, planned and researched a proposed venture; such behaviour suggests a significant role for variables unobserved in the panel surveys.
4. Testing strategy
4.3
55
Fixed-effect estimators
Let us maintain for now the assumption that wit is uncorrelated with the firm-specific time-variant it , but allow wit to be correlated with the firm-specific time-invariant ηi : E[wit ηi ] 6= 0; E[wit it ] = 0. In this case, a pooled estimator will no longer produce consistent estimates of γˆ2 ; in effect, measured values of wit will proxy for the unobserved ηi . However, “[t]he availability of panel data allows us to consistently estimate treatment effects without assuming ignorability of treatment and without an instrumental variable, provided the treatment varies over time and is uncorrelated with time-varying unobservables that affect the response” (Wooldridge 2002, 637). Thus, the fixed-effects estimator can be used to estimate γ2 consistently, by eliminating the time-invariant unobservable ηi . If z˜ refers to a fixed-effect transformation of some variable z (so that z˜ ≡ z − z¯), Equation 4.2 becomes: ˜ it β + ˜it . y˜it = γ1 d˜t + γ2 w ˜it + x
(4.3)
This equation therefore provides the identifying relationship for estimating the effect of legal status upon continuous variables (that is, upon the overdraft limit, the amount of short-term debt and the amount of long-term debt). However, for estimating the effect of legal status upon whether a firm has an overdraft — a binary variable — it is appropriate to use the fixed-effect logit estimator. It relies upon the following assumptions (see Wooldridge (2002, 458 & 482-492)). Assumption 1: Logistic functional form:
Pr(yit = 1 | xi , ηi ) = Λ(xit β + ηi ) =
exp(xit β + ηi ) . 1 + exp(xit β + ηi )
(4.4)
Assumption 2: Conditional independence of outcomes: yi1 and yi2 are independent conditional upon (xi , ηi ).
4. Testing strategy
56
Thus, in the fixed-effects logit context, Equation 4.2 is reformulated as:
Pr(yit = 1 | dt , wit , xit , ηi ) = Λ(γ0 + γ1 dt + γ2 wit + xit β + ηi ).
(4.5)
The fixed-effects logit estimator is the maximum-likelihood estimator derived from this specifying equation. The estimator maximises the joint distribution of yi ≡ (yi1 , yi2 ) conditional upon all ηi , (yi1 + yi2 ) and all explanatory variables. It operates upon observations for which yi1 + yi2 = 1; observations for which yi1 + yi2 = 0 or yi1 + yi2 = 2 “cannot be informative for β. . . because these values completely determine the outcome on yi ” (Wooldridge 2002, 491). That is, the fixed-effects logit estimator estimates the effect of legal status for those firms which either gained or lost their overdraft facility between the FACS and ICA surveys.
4.4
Robustness
To this point, we have assumed that changes in legal status are not affected by timevarying unobservables. Given the timing of the surveys used (discussed in Chapter Five), this is a concern that ought to be pursued. The problem we face is this. The legal change was imposed exogenously. Therefore, were we able to measure a sharp ‘before/after’ discontinuity, we could confidently interpret the estimated effect of changing legal status as a causal one; that is, we could be confident that the only effect being measured were that of the initial exogenous change in legal status. However, we do not have such a sharp measure: the second survey (ICA) was not conducted until 2004, three years after the legal reform. Thus, we must be concerned that, potentially, change in legal status is endogenous. Over any significant period of time — for example, the time between the implementation of the new law and the second round of the survey — one would expect there to be movement in firms’ fortunes: some growing and others contracting. Further, one would expect some of that movement to be reflected in
4. Testing strategy
57
characteristics observable to the bank but not directly measurable. Again, the example of management quality is apposite. Changes in the quality of a firm’s management between the FACS and ICA survey are likely to have been perceived by the firm’s bank, but are not recorded in the survey; further, a firm improving the quality of its management is presumably more likely to choose SA status over SARL status. If that is the case, we would anticipate an estimate of the effect of legal status to be biased.
Proxies There are two complementary strategies for dealing with this potential problem. First, we can control for observable changes in firm performance, on the assumption that changing “firms’ fortunes” are likely to correlate with other observed variables; thus, we would expect growing firms to increase the value of their capital and the number of their employees, and would expect the converse for those firms that are contracting. In effect, we use changes in other observed variables as proxies for changes in general firm performance. It is important to emphasise that, in doing so, we do not claim generally that the estimate of the effect of legal status in that relationship is a better estimate than that provided by the uncontrolled relationship; after all, it is also possible that changes in legal status cause changes in observable performance variables (that is, assets, employees, etc). Our ‘preferred regression’ shall remain, in each case, the uncontrolled fixed-effects estimate — however, whether the sign and significance of that estimate is preserved when controlling for other performance changes will be an important indication of the robustness.
Instruments However, even when controlling for such proxies, we may still fear for endogeneity; it may be that some changes in firms’ fortunes are not completely reflected in changes to other variables. Therefore, as a second robustness test, it is valuable to instrument
4. Testing strategy
58
for the change in firm status. The reason for instrumenting the change in firm status is clear when the fixed-effect specification is rewritten in terms of first differences (an alternative transformation to eliminate the time-invariant unobservable):
∆yit = γ1 + γ2 ∆wit + ∆xit β + ∆it .
(4.6)
Writing the equation in this way illustrates the concern: in order for γ2FE to be a consistent estimator, we must believe that E(∆wit ∆it ) = 0. However, this may not be reasonable. We specify, then, the following instrumenting equation, describing the determinants of a change in legal status:
∆wit = α1 + α2 z1i + α3 ∆z2it + ∆vit .
(4.7)
This equation allows that the change in legal status is explained both by the change of the vector z2i and by the level of the vector z1i at time t. (This is expressed in general terms; it will be argued shortly that the most appropriate method in the present context is to use only a vector z1i .) Without loss of generality, this then implies that:
wit = α0 + α1 dt + α2 dt z1i + α3 z2it + λi + vit .
(4.8) 0
Thus, the vector of (valid and informative) instruments is Zit =
dt z1i
z2it
.
Equation 4.6 can therefore be estimated by using two-stage least squares, instrumenting for ∆wit using Equation 4.7; the consequent estimator γˆ2IV will be consistent. Testing for endogeneity with a binary dependent variable Two-stage least squares may be used for instrumenting in the case of the continuous dependent variables. However, it clearly does not apply for the case of the binary dependent variable (existence of an overdraft). Instrumenting in that case is not a
4. Testing strategy
59
trivial exercise; because our regression specification is fixed-effect logit (rather than, for example, OLS), there is no clear method of instrumenting. The closest result is a methodology developed to test for exogeneity in other limited dependent variable models. The methodology was originally developed for the tobit model by Smith and Blundell (1986), and extended to the probit model by Rivers and Vuong (1988). The method is implemented as a two-stage estimation, assuming the specification of a primary relationship and an instrumenting relationship (Rivers and Vuong 1988, 353): 1. Regress the relevant explanatory variable (that is, the variable suspected of endogeneity) upon a vector of instrumental variables and upon all of the other explanatory variables in the primary relationship; 2. Take the least squares residuals from the first regression and include them as an explanatory variable in the primary regression. A test for their significance in that regression is a test of the hypothesis that the relevant variable is endogenous. Presently, we have the fixed-effect logit specification in Equation 4.5 and the instrumenting specification in Equation 4.8. To implement the Smith-Blundell methodology, we add the vector of explanatory variables to the second equation and express with a single error term:
wit = α0 + α1 dt + α2 dt z1i + α3 z2it + xit θ + uit .
(4.9)
A fixed-effect logit analogue to the Smith-Blundell test is therefore a test of the significance of π in the following regression:
Pr(yit = 1 | dt , wit , xit , u ˆit , ηi ) = Λ(γ0 + γ1 dt + γ2 wit + xit β + πˆ uit + ηi ),
where u ˆit is the vector of residuals from an OLS regression on Equation 4.9.
4. Testing strategy
60
To be clear, unlike the probit and tobit cases, there is no settled theoretic justification of the Smith-Blundell methodology in the logit case. However, given the similarity of the probit and logit tests, it is argued that the methodology is nonetheless useful in providing some insight into the endogeneity problem. Further, it appears to be the best approach available. (The alternative approach, of course, would be to use a probit form for the original specification and rely upon the result in Rivers and Vuong (1988); however, as explained, this is not possible because there is no fixed-effect probit estimator.) The methodology has intuitive appeal: if wit and yit are both driven by a common unobserved process (for example, changes in management quality), that process will be captured in u ˆit and, consequently, will be significant in the estimating equation for yit . It is important to emphasise that this provides a test for endogeneity rather than a method of correcting for it; thus, though our concern is to use the SmithBlundell methodology to test for endogeneity of legal change, the basic specification shall remain the preferred regression if that test is not significant.
Choice of instruments We seek, as always, a vector of instruments that is both valid and informative; a vector whose impact upon a firm’s change in bank overdraft status is assumed to operate only through its impact upon the firm’s change in legal status. (The specific variables used will be outlined in Chapter Six; for now, we merely describe the type of variables used.)
It was explained in Chapter Three that the change in legal status is likely to have imposed a differential change in cost across different qualities of firms. Recall that, in Figure 3.8, we imposed for simplicity a parallel shift in the ‘firm indifference curve’, but it was suggested that a ‘shift and swivel’ of the curve would allow cost to change differentially across different firm types. In effect, our model captured the intuition that the effect of the ‘intention to treat’ (that is, the firm-specific effect of the legal change)
4. Testing strategy
61
differed systematically across firms, and that this systematic difference was exogenous. Therefore, a vector of important firm characteristics observed before the reform, interacted with a time dummy to reflect the differential effect of the reform, produces a valid and informative instrumenting vector. In effect, we proceed by choosing observable measures of firm quality prior to the reform as the vector z1i in Equation 4.8.
At first blush, it may seem that such a vector could not be valid (that is, that it would have an direct causal effect upon banks’ overdraft decisions, separate from its effect upon legal status). This concern is overcome by an identifying assumption about the behaviour of the market. It is assumed (as in Chapter Three) that, both before and after the reform, the market lay in some kind of equilibrium state; further, it is assumed that there was no other significant credit market shock between the surveys. Therefore, while observable firm characteristics prior to the change may have been relevant in determining the characteristics of the initial equilibrium (as Chapter Three allows), those characteristics could not have affected the change in equilibrium but through the differential impact of the legal reform. For this reason, firm characteristics prior to the reform, interacted with a time dummy to reflect the effect of that reform, are justified as a valid instrumental vector.
4.5
Identifying asymmetric information by non-monotonicity
Hypothesis 2 — concerning the identification of asymmetric information by a concave relationship — requires a different econometric approach. This thesis proposes what, to the author’s knowledge, is a new empirical methodology for testing for information effects, distinct from incentive effects. In order to understand the significance of this econometric challenge, it is necessary to consider previous methodologies that have been used.
4. Testing strategy
4.5.1
62
Summary of related literature
The growing literature on the empirical estimation of contracting behaviour confronts a fundmental issue: how to distinguish asymmetric information effects (in the insurance context, ‘adverse selection’) from incentive effects (in the insurance context, ‘moral hazard’). This question is of direct relevance to the present research: if subsequent econometric analysis finds that ‘high status’ firms fare better than ‘low status’ firms in the credit market, this could be either (or both) because (i) ‘high status’ firms undertake enforceable obligations to act in a more transparent and responsible manner (the incentive effect) and/or (ii) because changing status reveals some otherwise unobserved information about firm quality (the information effect). Both explanations could be significant and valuable, but only the latter would speak to information asymmetry as a relevant cause of credit constraints. The literature has confronted this challenge in several ways, all of which (necessarily) involve identifying circumstances or ways in which information effects and incentive effects operate differentially upon the measured outcome.
An early example was Dionne and St-Michel (1991), which used a distinction between medical injuries (effectively, a difference between classes of insurance contracts) to differentiate information from incentive effects; because some injuries are easier for a doctor to diagnose (and hence more difficult for a patient to misrepresent), the differential effects of a change in insurance regulation upon reports of different types of accidents provides some insight into the relative effect of information and incentive effects. More recently, randomised experiments have been used. Both Ausubel (1999) and Karlan and Zinman (2006) differentiated information effects from incentive effects, in credit markets, by using randomised experiments. The key insight of both papers is that incentive effects and information effects operate at different times: thus, if subjects agree to a loan or credit contract under the promise of one interest rate, then
4. Testing strategy
63
randomly chosen participants have that rate lowered, the difference in outcome between those whose rate was lowered and those whose rate was maintained is a measure of the relative incentive effect alone, divorced from adverse selection effects. Karlan and Zinman (2006, 39) found “evidence of significant moral hazard, and weaker evidence of adverse selection” in their South African consumer loan market; in the US credit card context, Ausubel (1999) found broadly the opposite effect. Of course, to conduct a randomised experiment is a luxury lacking in many contexts, including that of Moroccan manufacturing firms. Recent work has used ‘quasi-experiments’ — specifically, legal reforms — to separate incentive from information effects; again, based upon the temporal distinction between those effects. It appears that only one paper has ever done this: Klonner and Rai (2006). Klonner and Rai study behaviour in Indian bidding ‘roscas’ before and after two policy shocks (the imposition of a bid ceiling in September 1993 and the relaxation of the ceiling in 2002) and identify “a significant adverse selection effect”, distinct from consequences of moral hazard.
4.5.2
Identification through non-monotonicity
The literature, then, suggests two dimensions along which incentive effects and information effects may operate differentially: across different types of contracts (Dionne and St-Michel 1991), and at different times in the contracting process (Ausubel 1999, Karlan and Zinman 2006, Klonner and Rai 2006). This thesis suggests a third dimension for identification: across different types of contracting agents. This approach is not suggested as a replacement for the two approaches outlined, but as an additional methodology for situations where — as here — neither of the other approaches can be used. (To be clear, we cannot use either approach because (i) we do not have significantly different classes of contracts, and (ii) the data does not distinguish post-contract and post-performance, nor does the present quasi-experiment provide a basis for arguing that contractual terms were unexpectedly changed post-agreement.)
4. Testing strategy
64
The key insight for the suggested methodology relies upon two identifying assumptions: (i) the probability density function of agent characteristics is non-monotonic, and (ii) incentive effects vary monotonically across agent characteristics. If we accept these two assumptions, we would expect a non-monotonic relationship of the same form of the probability density function between the effect of a treatment and underlying agent characteristics only if information asymmetries are significant to the relationship. That is, we would expect the non-monotonicity of the probability density function to be reflected in the outcome of the treatment only through information effects, rather than incentive effects. This insight can be understood by returning to the bivariate normal example of the model developed in Chapter Three. The two identifying assumptions were clearly met in developing that example. First, the example assumed a bivariate normal distribution of observable and unobservable quality — thus, the pdf was clearly non-monotonic (ie concave). Second, the incentive effects — captured by the function T (Qo , Qu ) — were not allowed to vary non-monotonically across agents.1
Under that bivariate normal specification, we derived the expected value of Qu conditional upon Qo and the observed signal φ. Consider, then, the difference between inferred Qu for φ = H and φ = L. Because of the role played by asymmetric information, we obtain:
E(Qu |Qo = qo , φ = H) − E(Qu |Qo = qo , φ = L) (4.10) σu ∗ µu +ρ σo (Qo −µo )−Qu √ φ p σu 1−ρ2 . = σu 1 − ρ2 · u u ∗ ∗ µu +ρ σ µu +ρ σ σo (Qo −µo )−Qu σo (Qo −µo )−Qu √ √ Φ · 1−Φ 2 2 σu
1
1−ρ
σu
1−ρ
Indeed, in the simplifying example, we set T (Qo , Qu ) = 0 for illustrative purposes. However, even if we had not done this, it would not be reasonable to imagine T (·) varying non-monotonically across agents — for example, it would not be reasonable to expect the incentive effects of the SA form to be higher for firms of ‘moderate’ quality than for firms of ‘lower’ and ‘higher’ quality.
4. Testing strategy
65
This is clearly a non-monotonic relationship; it is concave in Qo . In the simple model developed previously, the expectation of unobservable quality was relevant only for a binary decision: Lend or Refuse. However, it is not unreasonable to imagine that, in a real-world context, bank lending may depend in a continuous way upon bank inference of unobservable quality; that is, banks may lend more to firms perceived to be ‘better’, somewhat less to firms perceived to be ‘almost as good’, and so forth. In that case, a firm migrating from SA status to SARL status could be expected to lose a differential amount depending upon the bank’s initial assessment of its quality, and this amount would vary non-monotonically with the bank assessment. This is because, as the bivariate example illustrated, the SA signal is relatively more informative for firms at the extremities of the distribution than for firms closer to the middle.
Figure 4.1 shows this difference for the bivariate normal case; it plots the relationship shown in Equation 4.10. The relative ‘gain’ from migrating from SA to SARL status is negative and concave in observed quality (Qo ).
Fig. 4.1: A non-monotonic loss in the bivariate normal case
4. Testing strategy
66
Thus, if we can identify a significant concave relationship in the effect of the migration against bank’s initial perception of firm quality, we can conclude that information asymmetry was significant in the market and that the migration served a significant ‘information revelation’ function. Of course, unlike the alternative approaches outlined earlier, this approach does not promise a quantitative estimate of the relative importance of information effects over incentive effects. However, it does suggest a new methodology for testing whether information effects are significant, something that the alternative approaches cannot do with data of this form.
4.5.3
Identifying a concave relationship
So how, then, to seek a concave relationship in the present data? Two problems are paramount: choice of variables and choice of econometric technique. First, it is difficult to find obvious candidates for the relevant variables. The theoretical model predicts a non-monotonic relationship between bank’s inference of unobserved quality against bank’s initial assessment of observed quality among firms migrating from SA to SARL status; it is not obvious how this translates into particular variables. Given the results earlier in this chapter, the most obvious candidate to proxy for a change in the bank’s inference of unobserved quality must be the change in bank overdraft limit. As was shown earlier, the bank overdraft is not complicated by interacting considerations of supply and demand (unlike bank lending). As a proxy for the bank’s initial assessment of observed quality, we use the level of the bank overdraft limit at the time of the FACS survey. Second, it is vital to choose an appropriate econometric technique. It would be tempting to test for a concave relationship simply by adding a concave term to a regression (for example, a quadratic term). However, our model does not demand any particular functional form for the concave relationship, and we do not want to impose such a form a priori.
4. Testing strategy
67
For these reasons, we must turn to a non-parametric approach. The most appropriate method is kernel regression, which involves fitting a smoothed function across a series of points. In our case, we consider the set of firms migrating from SA to SARL status; for those firms, we use kernel regression to map a smooth relationship between the change in bank overdraft limit against the initial bank overdraft limit. (Specifically, we use the Nadaraya-Watson estimator: see Nadaraya (1964) and Watson (1964).) If that regression produces a concave estimate, we have prima facie evidence against the perfect-information null hypothesis. (In this context, a ‘concave estimate’ is taken to mean a kernel regression function that is globally weakly concave.)
4.5.4
Significance testing
Given the identifying assumptions used, finding such a concavity result — at some bandwidth — is necessary for even considering the issue of significance; if no concave function can be found, we know immediately that we cannot reject the perfectinformation null hypothesis. However, if a concave estimate is found, it is necessary then to test for its significance.
From the outset, it is important to emphasise what is being tested. We are not testing for the significance of any particular data point (or set of points) in relation to the regression function; thus, the confidence intervals produced by the kernel estimation are not directly useful for this purpose. Rather, our concern is to test for the significance of the conclusion that, as a whole, the data produces a concave kernel estimate function. We may specify the null and alternative hypotheses as follows. Denoting the initial bank overdraft limit as x and the change in overdraft limit as y, we have:
H0 : There is no relationship between x and y. H1 : y is a concave function of x.
4. Testing strategy
68
Clearly, H0 may be interpreted as the ‘perfect-information’ null hypothesis; if we cannot reject H0 , we cannot reject the null hypothesis that asymmetric information has no role in the market. Conversely, under the identifying assumptions used, H1 is equivalent to what was earlier specified as Hypothesis 2.
Given that we are forced to rely upon a non-parametric approach for estimation, it is not possible to derive a closed-form result to test the significance of that estimate. Instead, we may test for significance by obtaining imputed p-values through “simulating the null”: by simulating scenarios where the null hypothesis holds to learn how often we would expect the observed result to obtain. More specifically, the methodology proceeds as follows. • Take the observed values of X and Y . Randomise the matching of Y with respect to X (that is, so that Yi is matched with Xj , where i is not necessarily equal to j). Denote the new ordering of Y as Y˜ . By construction, Y˜ is now independent of X, though preserves the original distribution of Y . • Using an imposed bandwidth, run a kernel regression of Y˜ on X. Denote the resulting kernel estimates as yˆ˜. • Repeat the random matching and regression an arbitrary number of times, using the same bandwidth. Denote by c the number of regressions that produce a weakly concave kernel estimate function yˆ˜ of x (that is, a function that increases initially and turns exactly once). Denote by T the number of regressions. • For an imposed significance level p, Reject H0
for c ≤ p.T ;
Do not reject H0
for c > p.T .
(That is, the value
c is the simulated estimate of the p-value for the alternative T
4. Testing strategy
69
hypothesis, conditional upon the observed data.)2 As this outline intimated, there remains an important aspect of arbitrariness in the process: the selection of the bandwidth. The bandwidth is a smoothing parameter, such that a larger bandwidth produces a smoother function; thus, as the bandwidth tends towards zero, the kernel estimate tends to the underlying data, while as the bandwidth tends to infinity, the kernel estimate tends to the mean of the underlying data: H¨ardle and Linton (1994, 2309). It will therefore be necessary, in testing for the significance of a concave relationship, to consider how the imputed p-value varies as the bandwidth changes. This is an issue best considered by tangible application; it is therefore left to Chapter Six.
4.6
Conclusion
The previous chapter concluded by proposing two hypotheses to be tested; this chapter has outlined and justified the empirical methodology necessary to do so. By appealing to the concept of a treatment effect and imposing a linear functional relationship, the chapter justified the use of fixed-effect and first-difference estimators as the appropriate methodology for testing the variations on Hypothesis 1. Second, the chapter sought to make a methodological contribution by proposing a test of concave impact as a test of asymmetric information. A specific methodology for doing this — including for significance testing — was proposed. The following two chapters now implement these methods. Chapter Five briefly summarises and describes the data used, while Chapter Six reports the regression results.
2
It is important to note this conditioning: if we were not conditioning upon the observed data, we could simulate the null by regressing Y on a set of random numbers X drawn from some distribution. However, that approach requires an arbitrary assumption about the distribution of the null; by conditioning upon the observed data, we implicitly require the explanatory and dependent variables to take the same distribution as the observed data.
5. DATA
5.1
Data sources
The empirical analysis in this thesis draws upon three sources of data: the annual census of manufacturers, the Morocco FACS survey and the Morocco ICA survey. The sources are described in Fafchamps and El Hamine (2005); this outline summarises that description.
The census of manufacturers is conducted annually in Morocco by the Ministry of Commerce, Industry and Telecommunications (‘MCIT’). It has run since 1985, and its coverage is almost universal: “Only the smallest and most informal of firms, such as part-time home-based craft, are likely to escape the scrutiny of the Ministry” (Fafchamps and El Hamine 2005, Ch.2). For present purposes, the census is used to provide a description of the migration from the SA to SARL corporate form (in Chapter Two), and to provide the year of each firm’s creation.
The FACS (‘Firm Analysis and Competitiveness’) Survey was conducted in late 2000 as a collaborative venture between the MCIT, the World Bank and Oxford University. The sample was drawn by unstratified random sampling across six regions and seven production sectors. The FACS survey excluded firms with fewer than 10 employees. A total of 859 firms were interviewed.
The ICA (‘Investment Climate Assessment’) Survey was conducted in 2004. It was
5. Data
71
designed as a follow-up to the FACS survey and was again a collaborative project of the MCIT, the World Bank and Oxford University. Significant portions of the FACS survey — including the entire accounts section — were replicated in the ICA survey. A total of 746 firms was interviewed, 546 of which had been interviewed for the FACS survey.
Both the FACS and ICA surveys included detailed questions about firms’ recent financial statements; thus, though the surveys occurred only in 2000 and 2004, balance sheet variables are available — in some cases — from 1997 through to 2002. The data analysed here is the subset of firms in the FACS-ICA panel that were recorded as being either SA or SARL form in both the FACS and ICA surveys. This leaves a panel of 512 firms. Of those firms, 511 also appeared in the census.
5.2
Descriptive statistics
Tables 5.1 and 5.2 summarise the variables of interest: Table 5.1 refers to variables analysed in the two-period panel, while Table 5.2 refers to the extended panel formed by using balance sheet data. Most of the variables are self-explanatory; a few need clarifying. STdebt is a measure of short-term debt, and is constructed by summing the balance sheet measures of treasury papers (tr´esorerie-passif ) and recorded ‘other short-term debt’; LTdebt uses measured debts of financing (dettes de financement). The variable age records the firm age in the year 2000, as reported in the census data.
All of the non-negative continuous variables (ie all variables but bank odraft, ica, legstat, NetResultFACS and RetEarnFACS ) are distributed approximately lognormal; thus, to ensure that relevant relationships are not unduly identified by outliers, they enter regressions in log form. To overcome the problem of taking the log of a zero value, the specific functional form used is f (y) = log(y + 1).
5. Data
72
Tab. 5.1: Summary of variables: Two-period panel Variable Dependent variables bank odraft bank odraft amtT STdebt LTdebt Explanatory variables gross assets ica (dt ) legstat perm employees sales worth LandBuild worth MachEquip Interacted variables age × dt gross assetsFACS × dt NetResultFACS × dt perm employeesFACS × dt RetEarnFACS × dt
N
Mean
FACS S.Dev.
500 509 508 510
0.80 2435.83 10058.61 3363.20
0.40 8316.49 26213.32 15722.89
510 512 512 510 512 507 511
41472.28 0.0 0.37 131.01 27342.39 21472.00 16215.81
96190.45 0.0 0.48 221.26 58895.83 143849.9 65623.93
511 510 512 510 406
0.0 0.0 0.0 0.0 0.0
0.0 0.0 0.0 0.0 0.0
Max.
N
Mean
ICA S.Dev.
0 0 0 0
1 115000 322238 239618
512 510 406 406
0.74 2468.50 10680.65 3353.13
0.44 7610.69 27576.99 13367.34
0 0 0 0
1 100000 361112 164000
17 0 0 2 195 0 29
858780 0 1 3064 703940 2000000 1000000
510 512 512 512 511 471 509
48815.53 1.0 0.19 135.26 30142.57 15646.25 18824.9
109472.00 0.0 0.39 201.81 62492.92 61851.8 60116.7
225.04 1 0 10 2.3 0 1
915578.8 1 1 2056 630183.4 970000 700000
0 0 0 0 0
0 0 0 0 0
511 510 512 510 406
16.8 41472.28 1213.77 131.01 279.76
12.9 96190.45 7015.47 221.26 10483.93
1 17 -20812 2 -153274.6
100 858780 74047 3064 115415.5
Min.
Min.
Max.
5. Data
73
Tab. 5.2: Summary of variables: Five- and six-period panel Variable LTdebt
STdebt
gross assets
legstat
perm employ
sales
Year 1997 1998 1999 2000 2001 2002 1997 1998 1999 2000 2001 2002 1997 1998 1999 2000 2001 2002 1997 1998 1999 2000 2001 2002 1998 1999 2000 2001 2002 1998 1999 2000 2001 2002
N 424 434 434 347 362 350 424 434 433 346 361 350 429 434 434 429 429 434 436 436 436 436 436 436 435 434 419 435 436 435 436 435 435 436
Mean 3275.19 3326.87 3387.18 3259.12 3176.47 3227.77 10170.69 10771.48 11022.19 12721.35 11151.05 11463.97 40164.42 42982.37 45241.24 48731.12 50043.11 52608.91 0.83 0.79 0.71 0.49 0.28 0.25 133.27 139.29 135.40 140.82 139.55 31101.22 30073.35 32965.75 34271.67 32198.45
S.Dev. 13427.63 14778.52 16258.7 15100.16 14249.36 13837.16 26936.66 26932.43 28017.27 32444.34 27272.88 29135.54 88269.55 94053.24 100869.6 118767.4 108064.0 114147.0 0.38 0.41 0.46 0.50 0.45 0.43 212.94 235.42 202.48 210.18 205.53 61826.08 62502.97 95944.97 93495.74 65666.22
Min. 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 17.0 575.82 748.85 877.84 0 0 0 0 0 0 0 2 0 10 10 0.0 195.0 12.15 64.44 2.32
Max. 178012.0 174698.0 239618.0 172039.1 156436.5 164000.0 377084.0 342741.0 322238.0 363286.3 311307.0 361111.9 759870.0 830632.0 858780.0 1329612.0 944451.4 915578.8 1 1 1 1 1 1 3004 3064 1800 2056 2056 551504.0 703940.0 1439393.0 1496116.0 630183.4
Survey FACS FACS FACS ICA ICA ICA FACS FACS FACS ICA ICA ICA FACS FACS FACS ICA ICA ICA Census Census Census Census Census Census FACS FACS ICA ICA ICA FACS FACS ICA ICA ICA
6. ECONOMETRIC RESULTS
This chapter forms the culmination of our enquiry: it tests the hypotheses outlined in Chapter Three using the techniques justified in Chapter Four. The hypotheses are taken in turn.
6.1
Hypothesis 1: Choosing SA rather than SARL causes banks to be more willing to provide credit 6.1.1
Hypothesis 1.1: Concerning provision of an overdraft facility
This section uses the dependent variable bank odraft, a dummy variable recording whether or not the firm has an overdraft facility with its bank, to test whether those firms that changed legal status had a significantly higher or lower chance of having a bank overdraft.
The results from the fixed-effect logit are reported in Table 6.1 (page 75). The preferred regression is regression (1). Regressions (2) and (3) add various proxies for firm growth (entering in log form) as controls; those regressions seek to control for potential endogeneity arising from differences in firm performance. The regressors in regression (3) are relatively straightforward (ie number of permanent employees, sales, value of machinery and equipment, value of land and buildings and gross assets); those in regression (2) deserve explanation. Regression (2) uses as proxies the level, prior to the reform, of two variables anticipated to explain future firm growth: retained earnings
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
N Number of groups R2 χ2 statistic
IV residuals
Gross asset value (gross assets)
Value of land & buildings (worth LandBuild)
Value of machines & equipment (worth MachEquip)
Value of sales (sales)
No. of permanent employees (perm employees)
236 118 0.078 12.826
192 96 0.137 18.189
(0.00006)
0.00007
Net result (interaction) (NetResultFACS × dt )
(0.0001)∗
(0.236)∗
-0.344 -0.0003
(0.199)∗
-0.346
(0.805)∗∗
Proxies 1 (2) 1.834
Retained earnings (interaction) (RetEarnFACS × dt )
Time dummy, dt (ica)
(0.647)∗∗
Primary (1) 1.417
212 106 0.087 12.796
(0.478)
0.395
(0.138)
-0.053
(0.175)
-0.192
(0.212)
-0.145
(0.416)
58 29 0.405 16.295
(0.586)
-0.378
-0.336
(0.243)
(1.185)∗∗
‘IV’ subsample (4) 2.436
-0.226
(0.698)∗∗
Proxies 2 (3) 1.581
Tab. 6.1: Fixed-effect logit: Impact of legal status upon bank overdraft
Whether the firm has an overdraft (bank odraft): yes ↔ 1; no ↔ 0 Legal status (legstat): SA ↔ 1; SARL ↔ 0
58 29 0.444 17.83
(2.554)
2.999
(1.298)
-1.625
(2.293)
‘IV’ (Smith-Blundell) (5) 0.137
6. Econometric results 75
6. Econometric results
76
and annual net result (ie profit). Each level is then interacted with the time dummy. In effect, this seeks to explain the change in bank overdraft status by the level of profit and retained earnings prior to the reform. (Because these variables both allow negative values, neither enters in log form.) The estimated effect of legal status is positive and significant at the 95% level under all specifications. Indeed, the effect is estimated as larger when including the proxies. To be clear, as emphasised earlier, this is not a claim that the true effect exceeds that in regression (1); rather, it provides some confirmation that the effect in regression (1) is indeed causal.
Consider, then, the second test for robustness: the Smith-Blundell test for endogeneity, outlined in Chapter Four. The vector of instruments used — foreshadowed earlier — are observable determinants of firm quality prior to the legal reform (that is, measured in the FACS survey), interacted with a time dummy (so that, effectively, they are recorded as taking on the value zero for the first round of the panel and their FACS value for the second round of the panel). The variables used are age (measured from the census), gross assets and number of permanent employees; all enter in log form. Table 6.2 (page 77) reports the instrumenting equation used for the Smith-Blundell methodology: a fixed-effect OLS regression of legal status upon the instruments. (Note that, since we use the methodology to instrument under the basic specification of regression (1) in Table 6.1, there are no additional explanatory variables that need to be added to the regression in Table 6.2.)
It is clear immediately that instrumenting presents difficulties in this context; even when using instrumental variables central to firm performance, the instrumenting regression is very weak for the total sample of firms (regression (1), Table 6.2). The reason is a nonlinear relationship between firm performance indicators and change in legal status. Because movement in firm status is almost exclusively from SA to SARL,
6. Econometric results
77
Tab. 6.2: Fixed-effect OLS: Instrumenting for legal status
Legal status (legstat): SA ↔ 1; SARL ↔ 0 Age IV (age × dt ) Assets IV (gross assetsFACS × dt )
All firms (1) -0.055
Firms starting SA (2) 0.069
0.027
0.158
(0.027)∗∗
(0.047)
(0.017)
(0.027)∗∗∗
Employees IV (perm employeesFACS × dt )
-0.040
0.008
Time dummy, dt (ica)
-0.131
-2.402
(0.021)∗
(0.125)
Constant N Number of groups F statistic F test of joint instrument significance
0.367
(0.035) (0.229)∗∗∗
1.000
(0.013)∗∗∗
(0.022)∗∗∗
1015 508 23.745 2.17
373 187 89.017 23.23
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
and because it is firms of ‘middle’ quality that are likely to change status (as the model in Chapter Three suggests), a simple linear specification does not provide a strong instrumenting relationship. Attempts to use a quadratic function (not reported here) do not substantially improve the result. However, the instrumenting relationship is strong when specified for the subset of firms starting as SA status; among this subset, firms of ‘higher quality’ are clearly less likely to change status. This is reported in regression (2).
Therefore, the instrumenting robustness tests must be run only on the subset of firms starting as SA status. This is clearly less satisfactory than being able to instrument across all observations; however, it nonetheless provides some indication of how the basic estimation changes under instrumenting. Thus, the consequent fixed-effect logit, with the residuals from regression (2) of Table 6.2 included as an explanatory variable, is reported as regression (5) in Table 6.1. Since the number of instrumented
6. Econometric results
78
observations is significantly smaller than the total sample, regression (4) repeats the specification of regression (1) on that subsample to provide a basis for comparison. As explained earlier, the method presents a test for endogeneity, not a correction for it; thus, the primary parameter of interest is the effect of the residuals. This is not significant; therefore, the Smith-Blundell method does not reject the null hypothesis that the legal change was exogenous.
We therefore reject the null hypothesis that legal status has no effect upon the likelihood of having a bank overdraft. This repeats the result first reported in Fafchamps and El Hamine (2005, Ch.3), whose Table 5 corresponds to regression (1) in Table 6.1; the present analysis extends that by the addition of controls and the Smith-Blundell test.
Given the finding of a significant result, it bears pausing to consider the magnitude of the estimate. It was suggested earlier that the appropriate statistic to estimate here is the average treatment effect, that is, the partial effect attributable to choosing SA status over SARL status (see Equation 4.1). However, in this context, where the appropriate estimator was the fixed-effect logit, it is simply not possible to estimate this statistic, if the relevant dependent variable is understood as being ‘the probability of a firm having an overdraft’. This is because the estimation provided no estimate for the unobserved time-invariant, ηi . As Wooldridge explains, “we cannot estimate average partial effects, as doing so would require finding E[Λ(xit β +ηi )], a task that apparently requires specifying a distribution for ηi ” (2002, 492).1 However, what can be estimated is the partial effect of legal status upon the log-odds ratio.
Given the definition of the logit function (Equation 4.4), some basic algebra produces, 1
Wooldridge actually uses the notation ci instead of ηi .
6. Econometric results
79
in the context of this model: log
Pr(yit = 1 | dt , wit , xit , ηi ) 1 − Pr(yit = 1 | dt , wit , xit , ηi )
= γ0 + γ1 dt + γ2 wit + xit β + ηi .
Therefore, given that wit is a dummy variable, the estimate γˆ2 can be interpreted as the increase in the log-odds ratio for an SA firm relative to the same firm choosing the SARL status. Using the estimate from the preferred regression in Table 6.1, we estimate that effect to be an increase in the log-odds ratio of 1.417. Thus, we find a significant increase in the odds ratio of exp(1.417) ≈ 4.14. The implication of this result is shown in Figure 6.1, which shows the relative effect of choosing SA over SARL status (plus-or-minus one standard deviation) against the no-effect counterfactual (that is, the 45-degree line), conditional upon the (hypothetical) probability of the same firm having an overdraft as an SARL.2 Fig. 6.1: Estimated relative effect of choosing the SA status
2
That is, by conditioning upon such a hypothetical probability, we overcome the problem of not being able to identify an actual partial effect upon the probability). If we here allow y1i to be the probability of firm i receiving an overdraft with SA status and y0i to represent the same probability where the firm has SARL
y0i status, then the previous reasoning implies y1i = exp(γ2 ) 1−y
0i
y0i 1 + exp(γ2 ) 1−y
0i
−1
.
6. Econometric results
6.1.2
80
Hypothesis 1.2: Concerning limits on the overdraft facilitiy
The previous section showed that choosing SA form over SARL form significantly improves the probability of a firm having a bank overdraft. In this section, we extend the analysis to consider the relative effect of choosing SA form on the limit of the overdraft provided, treating firms without an overdraft as having a limit of zero.
The relevant variable (bank odraft amtT) is a truncated variable, bounded below by 0. Thus, were we to analyse the impact of legal status upon the variable in a crosssectional context, the appropriate estimator would be the tobit estimator. However, we have a panel and, for the reasons outlined earlier, our concern is to use a fixed-effect estimator. To date, no appropriate fixed-effect tobit estimator has been derived; therefore, the appropriate estimator is simply a fixed-effect OLS. Since overdraft limit is distributed approximately log-normal, the f (y) = log(y + 1) transformation is used.
Table 6.3 (page 81) shows the results for the fixed-effect OLS. It shows that legal status is significant and positive, both in the preferred regression (regression (1)) and when controlling for proxies (regressions (2) and (3)). Legal status is not significant, however, when the regressions are run upon the subset of firms retaining an overdraft in both periods (not reported); this suggests that, while legal status does have a significant positive effect upon the overdraft limit, that effect operates through banks cutting the overdraft completely, rather than by mere variations in the overdraft limit.
Regression (5) in Table 6.3 is the fixed-effect IV regression, using regression (2) in Table 6.2 as the instrumenting regression; regression (4) in Table 6.3 repeats the primary regression upon the instrumented subsample. A comparison of regression (4) and (5) suggests that instrumenting removes an otherwise positive estimate on the effect of legal status. As discussed in Chapter Four, this suggests the possibility that the result in
1019 512 5.273 0.02
N Number of groups F statistic R2
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
5.131
Constant
Value of sales (sales)
Value of land & buildings (worth LandBuild)
5.321
811 406 3.58 0.034
968 509 1.769 0.027
(3.289)
-0.503
(0.233)
0.031
(0.095)
-0.043
(0.111)
-0.055
(0.321)∗
(0.309)
0.222
(0.182)
-0.183
Value of machines & equipment (worth MachEquip)
(0.186)∗∗∗
0.00004
(0.00002)∗∗
(0.00002)
-8.42e-06
(0.183)∗
-0.309
(0.366)∗∗
Proxies 2 (3) 0.876
0.541
(0.162)∗∗∗
(0.157)
-0.181
(0.399)∗∗
Proxies 1 (2) 0.848
Gross asset value (gross assets)
No. of permanent employees (perm employees)
Net result (interaction) (NetResultFACS × dt )
Retained earnings (interaction) (RetEarnFACS × dt )
Time dummy, dt (ica)
(0.343)∗∗
Primary (1) 0.811 (0.449)∗∗
369 187 7.231 0.074
5.739
(0.476)∗∗∗
(0.326)
-0.208
IV subsample (4) 0.950
Tab. 6.3: Fixed-effect OLS: Impact of legal status upon bank overdraft limit
Firm overdraft limit (bank odraft amtT) (No overdraft ↔ bank odraft amtT = 0) Legal status (legstat): SA ↔ 1; SARL ↔ 0
369 187
6.276
(0.875)∗∗∗
(0.507)
-0.491
(0.860)
Fixed-effect IV (5) 0.413
6. Econometric results 81
6. Econometric results
82
regression (1) is driven by some source of time-varying endogeneity not captured by the proxies in regressions (2) and (3). However, it is nonetheless argued that the no-effect null hypothesis ought to be rejected; that is, it is argued that the result in regression (5) should not be seen as undermining the significant positive result in regression (1). There are several reasons for this conclusion. First, the positive significant result holds when correcting, in regressions (2) and (3), for important proxies of firm performance; thus, any candidate endogenous process would need to have effects beyond what is captured in those measures. Second, as noted earlier, it is possible to instrument only for firms starting as SA; the lack of significance in regression (5) could be explained by the relatively small sample. Third, as Figure 2.2 (page 11) showed, the migration occurred at approximately the same time as the legal change; therefore, any suggested endogenous process would also need to have operated at approximately that same time to have the suggested effect. Finally, it bears emphasising that the preferred regression already controlled for time invariant unobservables; this substantially narrows the range of potential claims of endogeneity. Therefore, though instrumenting suggests the possibility of endogeneity, we nonetheless reject the no-effect null hypothesis.
Before proceeding, it is instructive to consider a summary of the general trends in bank overdraft limit. Table 6.4 (page 83) provides this comparison across the four combinations of legal status. It shows that, while there was a significant reduction in overdraft size for firms migrating from SA to SARL status, there was no significant change in other firms’ outcomes. Intuitively, this is a surprising outcome: whether one describes the relevance of legal status in terms of information asymmetry or incentive effects, it should follow that firms remaining as SA form under the new legal regime received higher overdraft limits as a consequence of their decision (that is, because they revealed themselves to be of a higher ‘quality’ and/or because of the additional governance and transparency requirements undertaken). That they apparently did not
6. Econometric results
83
suggests that banks may not have responded even approximately optimally to the legal reform; it may be, for example, that they merely continued to follow simple ‘rule of thumb’ lending formulae, without adequately considering either the likely information effects or incentive effects of the reform. Tab. 6.4: Difference in mean overdraft across migration categories
Variable of interest: log(bank odraft amtT + 1) FACS ICA Difference in means Legal status Freq. Mean Std. Dev Mean Std. Dev t-value p-value SA ↔ SA 87 7.277 2.913 7.052 3.108 -0.491 0.624 SARL ↔ SARL 314 4.640 3.007 4.568 3.195 -0.291 0.771 SA ↔ SARL 97 6.174 2.294 5.024 3.124 -2.922 0.004∗∗∗ SARL ↔ SA 9 7.034 1.188 5.967 3.563 -0.852 0.407 All firms 507 5.423 3.029 5.106 3.295 -1.620 0.106 Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
6.1.3
Hypothesis 1.3: Concerning short-term lending
As in the case of bank overdraft facilities, the model suggests that legal status should have a significant and positive effect upon the banks’ willingness to provide formal loans (whether long- or short-term). However, one important empirical issue must be acknowledged immediately in the loan case: unlike the overdraft case, lending outcomes reflect effects both of banks’ willingness to lend (ie loan ‘supply’) and firms’ desire to borrow (ie loan ‘demand’). This is a stark contrast to the overdraft case, where firms’ ‘demand’ can reasonably be assumed to be inelastic to all measures of firm performance (that is, we can reasonably assume that all firms would like an overdraft facility, and would like the largest overdraft limit possible). That is, aside from being an issue of general importance for understanding firm credit, studying overdraft access allows a study directly of banks’ willingness to supply; in contrast, studying actual bank lending conflates that effect with firms’ desire for loans.
The empirical separation of demand and supply effects is not trivial. One approach is
6. Econometric results
84
to estimate demand and supply directly by using firm responses about loan applications (see Bigsten, Collier, Dercon, Fafchamps, Gauthier, Gunning, Oduro, Oostendorp, Patillo, S¨oderbaum, Teal, and Zeufack (2003)); however, we do not here have sufficient data in both the FACS and ICA surveys about whether and why firms did or did not apply for loans. More generally, one could try to separate the effects by finding an appropriate instrument that is assumed to affect demand but not supply (or vice versa). However, for simplicity, the present research limits itself only to instrumenting for change in legal status, as explained earlier.
Two-period panel Table 6.5 (page 85) reports the fixed-effect regression of short-term debt. The preferred specification is regression (1), which indicates that legal status has a negative impact upon the level of short-term debt, significant at the 90% confidence level. Regressions (2) and (3), including proxies, preserve this result. (Indeed, in regression (3), the significance increases to 99%; it has already been explained that this does not support a claim that the true effect is more significant than that in regression (1).) Therefore, despite the hypothesis that legal status causes the provision of more short-term debt, it seems that the effect is significant and negative. It is unclear exactly how to interpret this result. Given the theoretical model and the results on provision of bank overdraft, it is difficult to account for the result through bank willingness to supply loans (that is, it is difficult to imagine banks being more willing to lend because the SARL status is observed). Alternatively, the result could reflect the SA status causing firms to demand less debt. This could be understood, for example, by the SA status enabling better access to a bank overdraft (as earlier results suggest), as well as to more generous terms of supplier credit or better access to equity finance. As suggested, it seems that the difficulty in separating demand and supply for bank lending impedes a clear appreciation of the effects of legal status upon banks’ willingness to provide short-
914 510 6.557 0.032
N Number of groups F statistic R2
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
7.787
768 405 2.568 0.028
-0.009
Value of sales (sales)
Constant
0.111
Value of land & buildings (worth LandBuild)
873 506 6.800 0.117
(1.900)
-0.189
(0.180)
(0.054)∗∗
(0.062)∗∗
(0.178)∗∗∗
(0.177)
0.149
7.790
337 187 4.032 0.052
8.713
(0.341)∗∗∗
(0.232)
0.205
0.172
337 187
8.922
(0.652)∗∗∗
(0.369)
0.065
(0.645)
(0.324)
(0.106)
Fixed-effect IV (5) -0.657
IV subsample (4) -0.448
-0.043
(0.215)∗∗∗
Value of machines & equipment (worth MachEquip)
(0.102)∗∗∗
(1.00e-05)
-2.31e-06
(8.37e-06)
8.71e-07
0.172
(0.101)∗
(0.221)∗
Proxies 2 (3) -0.698
0.548
(0.097)∗∗∗
0.200
(0.096)∗∗
(0.211)∗
Proxies 1 (2) -0.396
Gross asset value (gross assets)
No. of permanent employees (perm employees)
Net result (interaction) (NetResultFACS × dt )
Retained earnings (interaction) (RetEarnFACS × dt )
Time dummy, dt (ica)
Legal status (legstat): SA ↔ 1; SARL ↔ 0
Primary (1) -0.397
Tab. 6.5: Fixed-effect OLS: Impact of legal status upon short-term debt, FACS & ICA
Short-term debt (STdebt)
6. Econometric results 85
6. Econometric results
86
term debt. The challenge of better separating those factors — and hence of explaining the significant negative effect — is left for future research. For completeness, regression (5) reports a fixed-effect instrumenting regression, using the instruments set out in Table 6.2. The result is not significant. However, neither it is significant in the uninstrumented regression run upon the same observations (regression (4)); instrumenting therefore contributes little in this case.
Six-period panel In Section 4.2, we assumed this relationship:
yit = γ0 + γ1 dt + γ2 wit + xit β + ηi + it ,
(4.2)
with t ∈ {0, 1} and dt = t. We can now maintain the same imposed relationship in the context of a six-year panel by allowing t ∈ {0, . . . , 5}. Then Equation 4.2 becomes, in differences:
∆yit = γ1 + γ2 ∆wit + ∆xit β + ∆it .
(6.1)
In effect, this structure imposes (i) that there is a constant year-on-year time change (γ1 ), and (ii) that changing legal status has the same given effect in any year. These assumptions are reasonable for first exploration, particularly given the relatively small number of identifying observations and relatively short panel duration.
Table 6.6 (page 87) reports a first-difference OLS of short-term debt. Regression (1) is the preferred regression. Regression (2) controls by including change in gross assets. Regression (3) includes still more controls, but (because data on sales and permanent employees is not available for 1997) it runs from 1998–2002. For comparison, regression (4) is the same specification as regression (1), run on the subset of observations
1863 0.017 9.27e-06
N F statistic R2
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
0.07
(0.028)∗∗
1404 19.998 0.054
(0.032)
1849 164.014 0.151
0.009 (0.026)
(0.086)
0.024
(0.055)
0.003
0.706
(0.084)∗∗∗
(0.085)
Controls 2 (3) 0.026
0.008
0.794
(0.044)∗∗∗
(0.077)
(0.083)
Constant
Change in no. of permanent employees (∆perm employees)
Change in sales value (∆sales)
Change in gross asset value (∆gross assets)
Change in legal status (∆legstat): SA ↔ 1; SARL ↔ 0
Controls 1 (2) 0.054
Primary (1) 0.011
Tab. 6.6: First-differences OLS: Impact of legal status upon short-term debt, 1997-2002
Change in short-term debt (∆STdebt)
1404 0.282 0.0002
0.062
(0.032)∗
(0.088)
Controls 2 subsample (4) 0.047 6. Econometric results 87
6. Econometric results
88
used for regression (3). The effect of legal status is not significant under any of the specifications. Given the instrumenting strategy used — namely, to use variables’ levels prior to the reform as instruments for change — we do not instrument in the sixperiod context. Therefore, both by considering the two-period panel and the six-period panel, we do not reject the no-effect null hypothesis of a positive effect of legal status upon short-term lending.
6.1.4
Hypothesis 1.4: Concerning long-term lending Two-period panel
The analysis for long-term debt mirrors that for short-term debt. However, it bears noting immediately that, intuitively, we would anticipate that significant results are less likely to emerge in the long-term debt case. Long-term debt is a less important source of firm finance, both in its magnitude (see Table 5.2) and in its use; unlike shortterm debt, the median value of long-term debt (unreported here) is zero, both for the FACS and ICA surveys. Consider, then, the results. Table 6.7 (page 89) is perfectly analogous to Table 6.5, for the case of long-term debt; the preferred specification is again regression (1). Legal status is not significant under any specification.
Six-period panel Table 6.8 (page 90) is perfectly analogous to Table 6.6; it reports the effect of legal status upon long-term debt, using the six-period panel. Again, as in the case of shortterm debt, legal status is not significantly positive under any specification. Therefore, as in the case of short-term debt, neither the two-period nor the six-period panel allow any rejection of the no-effect null hypothesis. In short, we do not find significant effects of legal status upon either short-term or long-term firm debt, either under the simple FACS-ICA panel or using the extended panel. It is unclear whether this non-result reflects (i) flaws in the theory being tested, (ii) the conflation of firm demand with bank
916 511 0.895 0.004
N Number of groups F statistic R2
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
3.276
(0.189)∗∗∗
(0.201)
(0.187)
770 406 0.995 0.011
3.147
(0.203)∗∗∗
(0.00002)
-0.00002
(0.00002)
0.00003
0.205
0.196
(0.439)
(0.411)
Constant
Value of sales (sales)
Value of land & buildings (worth LandBuild)
Value of machines & equipment (worth MachEquip)
Gross asset value (gross assets)
No. of permanent employees (perm employees)
Net result (interaction) (NetResultFACS × dt )
Retained earnings (interaction) (RetEarnFACS × dt )
Time dummy, dt (ica)
Legal status (legstat): SA ↔ 1; SARL ↔ 0
Proxies 1 (2) 0.039
Primary (1) -0.140
875 507 0.726 0.014
(3.989)
-2.149
(0.378)
0.033
(0.114)
0.114
(0.131)
0.109
(0.374)
0.381
(0.371)
-0.066
(0.223)
-0.001
(0.448)
Proxies 2 (3) -0.162
338 187 0.7 0.009
3.517
(0.64)∗∗∗
(0.437)
0.476
(0.608)
IV subsample (4) 0.278
Tab. 6.7: Fixed-effect OLS: Impact of legal status upon long-term debt, FACS & ICA
Long-term debt (STdebt)
338 187
3.192
(1.217)∗∗∗
(0.691)
0.643
(1.201)
Fixed-effect IV (5) 0.603
6. Econometric results 89
Confidence: *** ↔ 99%, ** ↔ 95%, * ↔ 90%.
N F statistic R2
Constant
Change in no. of permanent employees (∆perm employees)
Change in sales value (∆sales)
Change in gross asset value (∆gross assets)
Change in legal status (∆legstat): SA ↔ 1; SARL ↔ 0
1855 0.997 0.001
(0.055)
(0.054)
1869 1.293 0.0007
0.013
0.016
(0.17)
1410 0.897 0.003
(0.065)
-0.016
(0.174)
0.084
(0.111)
0.045
0.013
(0.172)∗
(0.092)
(0.16)
(0.158)
Controls 2 (3) -0.301
0.041
Controls 1 (2) -0.212
Primary (1) -0.180
Tab. 6.8: First-differences OLS: Impact of legal status upon long-term debt, 1997-2002
Change in long-term debt (∆LTdebt)
1410 3.055 0.002
(0.063)
-0.014
(0.172)∗
Controls 2 subsample (4) -0.301 6. Econometric results 90
6. Econometric results
91
supply in the lending case (as discussed earlier) or (iii) a distinction between the fact of lending and the amount lent (analogous to the result suggested in Section 6.1.2). This, too, presents opportunities for further work.
6.2
Hypothesis 2: Signalling effects contribute to the relative value of choosing SA status
It was suggested in Chapter Four that the significance of signalling effects can be tested by considering whether a significant concave relationship exists between change in bank overdraft limit and the initial value of the overdraft limit; a methodology for doing so was suggested. In this section, we apply that methodology. As described, we use the change in the truncated overdraft limit variable among firms switching from SA to SARL (ie ∆bank odraft amtT) to proxy for the relative impact of legal status upon bank perceptions of firm quality. The initial overdraft limit (ie bank odraft amtTFACS ) is used to proxy the bank’s initial perception of firm quality. Specifically, since the variable is approximately distributed log-normal, a log transformation is used to ensure that the points are reasonably spread. (Since the variable may take the value zero, the specific transformation is, again, f (y) = log(y + 1).)
Under this specification, a weakly concave kernel estimate, turning once, obtains for all bandwidth values (up to two decimal places) from 0.85 to 1.38 (inclusive). It bears noting that there is no bandwidth that produces a convex kernel estimate. (For bandwidths less than 0.85, the function turns multiple times; for bandwidths greater than 1.38, the function decreases monotonically.) Figure 6.2 illustrates the concave kernel estimate produced for a bandwidth of 1.1; the following figure focuses upon the turning point.3 Table 6.9, then, reports the results of ‘simulating the null’ (as explained in Chapter Four) for various bandwidths in the relevant range. The imputed p-value is the propor3
Though it is not directly relevant — as explained in Chapter Four — both graphs plot for completeness a region of plus-or-minus 1.96 standard deviations from the estimate.
6. Econometric results
92
Fig. 6.2: Kernel regression of those firms switching legal status (Bandwidth: 1.1)
Fig. 6.3: Kernel regression of those firms switching legal status (Bandwidth: 1.1): Focusing upon the turning point. (Points plotted are the kernel estimate points, not the data points.)
6. Econometric results
93
tion of simulations producing a kernel estimate initially increasing and turning once (that is, the values reported in the second column); the value varies across the bandwidth range, but remains generally above 0.2 and always above 0.16. Figure 6.4 illustrates this. For completeness, we consider also dropping those firms not initially having an overdraft (that is, firms having bank odraft amtTFACS = 0); this reflects a concern that the earlier result may be driven by the inclusion of the ‘zeroes’, lying a distinct distance from the other points. In that case (whose specific results are omitted for brevity), we find a concave kernel estimate for bandwidths from 1.34 to 1.93; the imputed significance of that result ranges from 0.201 to 0.298. Tab. 6.9: Simulation results: 1000 valid simulations at varying bandwidths
Bandwidth 1.38 1.35 1.30 1.25 1.20 1.15 1.10 1.05 1.00 0.95 0.90 0.85
Turning points: Function initially increasing None One Multiple 0.203 0.188 0.029 0.204 0.218 0.040 0.178 0.207 0.043 0.157 0.218 0.038 0.119 0.224 0.102 0.097 0.242 0.136 0.086 0.263 0.123 0.054 0.254 0.179 0.030 0.260 0.191 0.025 0.218 0.242 0.023 0.207 0.280 0.012 0.162 0.313
Turning points: Function initially decreasing None One Multiple 0.244 0.312 0.024 0.224 0.283 0.031 0.227 0.304 0.041 0.178 0.356 0.053 0.131 0.331 0.093 0.086 0.298 0.141 0.060 0.293 0.175 0.034 0.286 0.193 0.030 0.284 0.205 0.027 0.258 0.230 0.019 0.195 0.276 0.017 0.192 0.304
Sum 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
Therefore, the suggested methodology of ‘simulating the null’ is unable to reject the null hypothesis that there are no information effects operating in the relevant credit market. Of course, this does not disprove the importance of asymmetric information, a factor that is repeatedly identified as a relevant issue by Moroccan policymakers and businesspeople. One reason that the hypothesis cannot be rejected may be the relatively small number of useful observations: the results in Table 6.9 are identified by 99 observations, while the results omitting the zeroes are identified by 90. Further re-
6. Econometric results
94
Fig. 6.4: Imputed p-values across varying bandwidth
search is needed to understand how additional assumptions may assist in identifying an information effect; for example, by conditioning upon other covariates, whether by a ‘matching’ approach or by imposing a more specific functional form. Additional consideration must also be paid to the choice of relevant proxy variables; it bears noting that some other candidate proxies for observable firm quality (for example, firm age) fail even to produce a concave kernel estimate.
Nonetheless, data limitations often prevent identification of different classes of contracts (as in Dionne and St-Michel (1991)) or exogenous post-agreement changes in contractual terms (Ausubel 1999, Karlan and Zinman 2006, Klonner and Rai 2006); the methodology suggested and applied here may form a useful additional method for identifying information effects.
6. Econometric results
6.3
95
Empirical conclusions
The empirical results outlined produce the following conclusions to the specified hypotheses.
Hypothesis 1: Choosing SA rather than SARL status causes banks to be more willing to provide credit. • Hypothesis 1.1: Upon overdraft facility: Reject the no-effect null hypothesis at the 95% level: legal status does have this effect. • Hypothesis 1.2: Upon overdraft limit: Reject the no-effect null hypothesis at the 95% level: legal status does have this effect. • Hypothesis 1.3: Upon short-term lending: Do not reject the null hypothesis. • Hypothesis 1.4: Upon long-term lending: Do not reject the null hypothesis.
Hypothesis 2: The value of choosing SA rather than SARL is due, in part at least, to signalling effects: Do not reject the perfect-information null hypothesis.
7. CONCLUSIONS
Summary This thesis has sought to make contributions theoretical, empirical and methodological. To the author’s knowledge, no theoretical model has considered credit constraints as arising from a continuous unobserved variable signalled by a discrete choice variable. As Chapter Three showed, a number of interesting and informative insights emerge from just such a model. First, a discrete signal of continuous firm quality juxtaposed credit constraints against non-performing loans; they were shown to be — literally — opposite sides of the same phenomenon. Second, the model drew attention to the relationship between the cost of the signal and the profitability of the signalling firm; credit constraints are not inherently caused by asymmetric information so much as they are caused by the inability of profitable firms to invest in the informative signal (and for non-performing loans, the converse). Extending the model to the bivariate case — where firm quality is partially but not completely observed — suggested further insights. First, the extension showed a way of considering the informativeness of a signal without introducing an exogenous source of ‘noise’: a signal is more informative if its cost varies more strongly with unobserved characteristics than with observed characteristics. Second, by modelling observed characteristics explicitly, the bivariate extension formed a natural vehicle for motivating empirical analysis; for example, it allowed explicitly that observably different firms (that is, firms with different observable qualities) are affected differently by changes in the cost of signalling. Indeed, the most interesting implication of this form was that, if asymmetric information is significant
7. Conclusions
97
in a credit market, a change in the cost of signalling should induce a non-monotonic relationship between change in credit outcomes and initial assessments of firm quality.
It also bears noting that the model developed here extends the existing literature in a simple framework. To be sure, this has its drawbacks — for example, there is no inherent role for credit-specific parameters like the interest rate or collateral. However, conversely, it allows the possibility that the model could readily be applied to different contexts. Indeed, the model could apply to almost any context in which an uninformed principal must infer an unobserved component of an agent’s quality on the basis of observable quality and a binary signal; it is not difficult to think of scenarios to which this could apply (for example, certification effects in the labour market, audit ratings of local government authorities, property titling in developing economies, etc).
Empirically, the thesis has sought to contribute the first econometric evaluation focussing specifically upon Morocco’s new company law. The insights are threefold. First, by confirming that legal status has a significant causal effect upon overdraft provision and overdraft limits, the research shows that Moroccan banks view as significant the difference in legal obligations imposed by the SA and SARL forms; consequently, this implies that the legal reform will have ongoing relevance to the credit market. (Further, in a more general sense, it suggests that firm-level econometric analysis may often need to consider legal status as an explanatory variable; the combined effect of the legal outline and empirical results in this research is to show that different legal forms can imply substantially different legal obligations, and can attract significantly different market treatments.) Further, the research suggests empirically that asymmetric information may not play a significant role in the Moroccan credit market, notwithstanding common views to the contrary. Of course, this last conclusion is driven by a non-result, so may attract revision upon analysis of more data or imposition of stricter identifying
7. Conclusions
98
assumptions.
Finally, the thesis sought a methodological contribution to the analysis of imperfect information, by suggesting and applying a new method of separating information and incentive effects. This involved seeking a concave relationship between the impact of an exogenous shock and the uninformed agent’s initial perception of the quality of the informed agent; the method of “simulating the null” was suggested to test the significance of this. Though this method failed to find significant evidence of asymmetric information in the present data, the method could potentially be applied to a wide variety of other contract contexts (eg labour markets) to test for information asymmetry independent of incentive effects.
Implications The findings of this research hold implications both for Moroccan banks and for Moroccan policymakers. For Moroccan banks, the research — particularly its emphasis upon the additional obligations undertaken by SA companies and the likelihood of a consequent ‘transparency gain’ conferred upon lenders — suggests the possibility that bank lending practices have not properly accommodated the legal change. The model emphasises that an optimising lender will respond to an exogenous legal change both by (i) accounting for the differential legal obligations, and — though this was not modelled directly — (ii) by conditioning both upon present legal status and previous status. A lender that merely adheres to its previous ‘rule of thumb’ lending rules will unnecesarily refuse profitable loans; a more sophisticated approach (as suggested) will produce a Pareto improvement in reducing credit constraint problems. To be sure, this research did not seek to identify such inefficient lending strategies — this is left as a potential avenue for further research — but it does suggest the possibility that Moroccan banks have not adequately responded to the legal reform. Organisations working
7. Conclusions
99
to assist the modernisation of Moroccan banks’ loan assessment technologies (for example, the International Finance Corporation) may find this a valuable point to make in dialogue with the banking sector.
It is difficult to see obvious implications for the Moroccan government; certainly, the research is not intended to imply that, because migration to the SARL status had a causal effect upon the loss of bank overdraft, the reform was a mistake. Indeed, to the contrary, the result that lenders condition upon legal status — and that this is apparently drive in large part by incentive effects rather than information effects — suggests that reforms to improve the quality of corporate governance ought, in the longer term, improve firms’ prospects of credit access. An additional insight is suggested by the model: if, despite the non-result here, policymakers believe that information asymmetries remain an important cause for credit constraints, it follows that there may be a role for policy in seeking to allow firms an increased number of verifiable signals. This reflects Spence’s (1973, 367) concern that “effective signaling depends not only upon the negative correlation of costs and productivities, but also upon there being a ‘sufficient’ number of signals within the appropriate costs range”. One obvious example — but by no means the only possibility — would be to allow firms to choose a variable but verifiable “minimum capital requirement” above some absolute minimum level. This may allow higher quality firms to distinguish themselves more effectively from lower quality competitors; in turn, as the theoretical model shows, this may assist in reducing both the incidence of credit constraints and of non-performing loans.
Future research This research suggests a number of avenues for further enquiry. Some have already been noted throughout; for example, further research may seek to separate demand and supply effects in short- and long-term bank lending, and may suggest means for
7. Conclusions
100
identifying the extent to which banks have optimally responded to the legal reform. But there remain further possibilities still. A third round of the panel data may soon be available; if so, this could provide an important basis for further testing of the results identified here, and for assessing the way that banks and firms have subsequently ‘learned’ about the consequences of the legal reform (for example, it may emerge that banks’ initial response was to follow overly simplistic lending rules, but that such rules have subsequently been revised). More adventurous research could move beyond the simple ‘lend/refuse’ framework to consider data upon other contractual terms, particularly collateral and interest rates. Morocco has infamously high average collateral pledges upon loans; the present results and research may provide a starting point for research into the nature and causes of that phenomenon.
Finally, beyond particular economic and econometric questions, the thesis illustrates again the value of studying markets and institutions in Morocco. Even leaving aside the many imperatives for effective policy in Morocco at the present time — that it is African, Islamic, modernising, rapidly integrating into European markets and so forth — the country provides a particularly valuable context for improving our understanding of market institutions in development. Morocco is at once a developing country and a modernising economy; a place confronting developing-economy market inefficiencies while having a legal system with clear and identifiable rules; and, perhaps most importantly, it is a system whose polity has been willing to experiment with radical institutional reform in the name of economic and legal modernisation. For all of these reasons, it is a context particularly apposite for better understanding market institutions and their relationship to the development process; a context promising much for future research.
BIBLIOGRAPHY
A LAMI , K. C. (1999): “Tribune libre: Pourquoi choisir la soci´et´e anonyme,” L’Economiste (accessed online: http://www.leconomiste.com/), 28 April 1999. AUSUBEL , L. (1999): “Adverse Selection in the Credit Card Market,” Working paper, available at http://www.ausubel.com/creditcard-papers.htm, Version of 17 June 1999. B ESANKO , D., AND A. T HAKOR (1987): “Collateral and Rationing: Sorting Equilibria in Monopolistic and Competitive Credit Markets,” International Economic Review, 28(3), 671–689. B ESTER , H. (1985): “Screening vs. Rationing in Credit Markets with Imperfect Information,” American Economic Review, 75(4), 850–855. (1987): “The Role of Collateral in Credit Markets with Imperfect Information,” European Economic Review, 31(4), 887–899. B IGSTEN , A., P. C OLLIER , S. D ERCON , M. FAFCHAMPS , B. G AUTHIER , J. W. ¨ G UNNING , A. O DURO , R. O OSTENDORP, C. PATILLO , M. S ODERBAUM , F. T EAL , AND A. Z EUFACK (2003): “Credit Constraints in Manufacturing Enterprises in Africa,” Journal of African Economies, 12(1), 104–124. B OLTON , P., AND M. D EWATRIPONT (2005): Contract Theory. The MIT Press. C HBANI I DRISSI , J. (1996): “Nouvelle loi sur la soci´et´e anonyme: La responsibilit´e des administrateurs et des commissaires aux comptes interpell´ee,” L’Economiste (accessed online: http://www.leconomiste.com/), 14 November 1996. C HO , I.-K., AND D. K REPS (1987): “Signaling Games and Stable Equilibria,” Quarterly Journal of Economics, 102(2), 179–222. D IONNE , G., AND P. S T-M ICHEL (1991): “Worker’s Compensation and Moral Hazard,” Review of Economics and Statistics, 73, 236–244. E L H AMMOUMI , A. (2005): “Contractualisons le droit des soci´et´es!,” L’Economiste (accessed online: http://www.leconomiste.com/), 17 May 2005. L’Economiste (1998a): “SA: Quel est le dead-line pour harmoniser les statuts?,” L’Economiste (accessed online: http://www.leconomiste.com/), 15 January 1998. (Author unnamed).
Bibliography
102
(1998b): “SARL: Plus soft de la SA, mais,” L’Economiste (accessed online: http://www.leconomiste.com/), 8 October 1998. (Author unnamed). (1999a): “La SA prise au pi`ege au Tribunal de Commerce,” L’Economiste (accessed online: http://www.leconomiste.com/), 23 November 1999. (Author unnamed). (1999b): “Pour la cr´eation d’une formule nouvelle de centre de formalit´es des enterprises,” L’Economiste (accessed online: http://www.leconomiste.com/), 25 November 1999. (Author unnamed). (1999c): “Soci´et´e anonyme: Report en pl´eni`ere au Parlement,” L’Economiste (accessed online: http://www.leconomiste.com/), 31 December 1999. (Author unnamed). (2000a): “Harmonisation des statuts des SA/SARL: Les premi`eres r´eactions au sursis,” L’Economiste (accessed online: http://www.leconomiste.com/), 3 January 2000. (Author unnamed). (2000b): “Soci´et´e Anonyme: Pas de report cette fois,” L’Economiste (accessed online: http://www.leconomiste.com/), 21 December 2000. (Author unnamed). (2005): “R´eforme de la SA: La refonte de la r´egime des valeurs mobili`eres s’impose,” L’Economiste (accessed online: http://www.leconomiste.com/), 28 July 2005. (Author unnamed). FAFCHAMPS , M., AND S. E L H AMINE (2005): “The Manufacturing Sector in Morocco,” Unpublished report for the World Bank, submitted to the Government of Morocco. G UENNOUNI , H. (2004): “Soci´et´es anonymes non cot´ees au Maroc: L’indispensable modernisation du cadre juridique,” L’Economiste (accessed online: http://www.leconomiste.com/), 2 November 2004. ¨ H ARDLE , W., AND O. L INTON (1994): Applied Nonparametric Methodsvol. IV of Handbook of Econometrics, chap. 38. Elsevier Science. JAFFEE , D., AND F. M ODIGLIANI (1969): “A Theory and Test of Credit Rationing,” American Economic Review, 59(5), 850–872. JAFFEE , D., AND T. RUSSELL (1976): “Imperfect Information, Uncertainty and Credit Rationing,” Quarterly Journal of Economics, 90(4), 651–666. K ARLAN , D., AND J. Z INMAN (2006): “Observing Unobservables: Identifying Information Asymmetries with a Consumer Credit Field Experiment,” Center for Global Development, Working Paper 109; available at http://www.cgdev.org/content/publications/detail/12329/, Version of 17 June, 2006. K LONNER , S., AND A. R AI (2006): “Adverse Selection in Credit Markets: Evidence from Bidding Roscas,” Working paper; available at http://www.arts.cornell.edu/econ/sklonner/, Version of 1 June, 2006.
Bibliography
103
L AZRAK , R. (2004): “La n´ecessaire mise a` niveau l´egislative,” L’Economiste (accessed online: http://www.leconomiste.com/), 5 May 2004. L OFTI , M. (1996): “Transformation des SA en SARL: Les formalit´es juridiques et fiscales,” L’Economiste (accessed online: http://www.leconomiste.com/), 19 December 1996. M AATOUK , J. (2002): “Amendes contre prison: L’impossible e´ quilibre du droit des soci´et´es anonymes,” L’Economiste (accessed online: http://www.leconomiste.com/), 8 March 2002. M ARTIN , D. R. (1999): Droit commercial et bancaire marocain. Soci´et´e d’Edition et de Diffusion Al Madariss, Casablanca. M EYER , M. (1991): “Learning from Coarse Information: Biased Contests and Career Profiles,” Review of Economic Studies, 58a(1a), 15–41. M OSSADAQ , F., AND M. O UDGHIRI (1995): “R´eforme de la soci´et´e anonyme: M Saa¨ıdi explique les innovations,” L’Economiste (accessed online: http://www.leconomiste.com/), 21 December 2005. M SALHA , M. (2005): Les Societes Commerciales en Droit Marocain. Imprimerie Najah el Jadida, Casablanca. NADARAYA , E. (1964): “On estimating regression,” Theory of Probability and its Applications, 10, 186–190. O SBORNE , M., Press.
AND
A. RUBINSTEIN (1994): A Course in Game Theory. The MIT
O UDGHIRI , M. (1998): “Ce qu’il y a a` retenir de la loi,” L’Economiste (accessed online: http://www.leconomiste.com/), 2 December 1998. O UDGHIRI , M., AND R. I KRAM (1999): “Soci´et´e anonyme: Amender d’abord la loi,” L’Economiste (accessed online: http://www.leconomiste.com/), 26 November 1999. R IVERS , D., AND Q. V UONG (1988): “Limited Information Estimators and Exogeneity Tests for Simultaneous Probit Models,” Journal of Econometrics, 39(3), 347– 366. S ALANI E´ , B. (1997): The Economics of Contracts. The MIT Press. S EN , A. (1999): Development as Freedom. Oxford University Press. S HAMAMBA , A. (1995): “Soci´et´e anonyme : L’effectivit´e du contrˆole, garant du succ`es de la r´eforme,” L’Economiste (accessed online: http://www.leconomiste.com/), 21 December 2005. (2002): ““La pluridisciplinarit´e des big-five est in´eluctable”: Azzedine Benmoussa, associ´e-fondateur de KPMG Maroc,” L’Economiste (accessed online: http://www.leconomiste.com/), 29 March 2002.
Bibliography
104
S LAOUI , M., AND M. L ECERF (2000): “SA: L’amendement aurait e´ t´e pr´eferable au report,” L’Economiste (accessed online: http://www.leconomiste.com/), 21 January 2000. S MITH , A. (1776): An Inquiry into the Nature and Causes of the Wealth of Nations. Oxford University Press, 1976 edition, edited by R.H.Campbell and A.S.Skinner. S MITH , R., AND R. B LUNDELL (1986): “An Exogeneity Test for a Simultaneous Equation Tobit Model with an Application to Labor Supply,” Econometrica, 54(3), 679–686. S PENCE , M. (1973): “Job Market Signaling,” Quarterly Journal of Economics, 87(3), 355–374. S TIGLITZ , J., AND A. W EISS (1981): “Credit Rationing in Markets with Imperfect Information,” American Economic Review, 71(3), 393–410. T OBIN , J. (1958): “Estimation of Relationships for Limited Dependent Variables,” Econometrica, 26(1), 24–36. WATSON , G. (1964): “Smooth regression analysis,” Sankhya Series A, 26, 359–372. W ETTE , H. (1983): “Collateral in Credit Rationing in Markets with Imperfect Information: Note,” American Economic Review, 73(3), 442–445. W OOLDRIDGE , J. (2002): Econometric Analysis of Cross Section and Panel Data.
8. APPENDIX: DETAILS OF THE LEGAL REFORM
8.1
An overview of Moroccan law
Under the 1996 Constitution of Morocco,1 the laws of Morocco are made by a bicameral legislature (comprising the Chambre des Repr´esentants and the Chambre des Conseillers).2 Members of the Chambre des Repr´esentants are elected for six-year terms by direct universal suffrage;3 members of the Chambre des Conseillers are elected for nine-year terms by representatives of various interest groups.4 Parliament votes upon proposed legislation,5 and legislation is generally enacted by each chamber adopting an identical text.6 A law adopted by the parliament is promulgated by the King within 30 days of its receipt by the government,7 and the government is then responsible for ensuring the implementation of the law throughout Morocco.8 The instrument used for royal promulgation is known as a ‘dahir’, a term that will be used repeatedly in the following discussion; it fulfils a similar role to that played in England by the Royal Assent. 1
2 3 4 5 6 7 8
Adopted on 13 September 1996 and currently in force. As subsequent discussion will outline, some of the constitutional steps relevant to the reforms considered here occurred prior to 13 September 1996; thus, the reform process was governed in its early stages by the previous text, the Constitution of Morocco of 4 September 1992. However, the following discussion is nonetheless a relevant overview of the general process. Broadly, the Constitution of 13 September 1996 strengthened parliamentary oversight and the rule of law, but did not significantly change the legislative process described here. Article 36, Constitution of Morocco. Article 37, Constitution of Morocco. Article 38, Constitution of Morocco. Article 45, Constitution of Morocco. Article 58, Constitution of Morocco. See the subsequent provisions of Article 58 for some exceptions. Article 26, Constitution of Morocco. Article 61, Constitution of Morocco.
8. Appendix: Details of the legal reform
8.2
ii
The new legal regime: a comparison between SA and SARL
A comparison between law 17-95 (regarding the SA) and law 05-96 (regarding the SARL) is a comparison between two very different legal regimes. Even before considering the differences in specific rights and obligations that the two statutes confer, it bears noting the fundamental difference in the underlying character of the two forms. “[T]he SA was designed as a joint stock company drawing upon public saving. It it thus distinguished very clearly from the SARL, a structure intended for companies of average size and/or of a family character, where the personality of the contributor is generally more important than the capital contributed.”9 (Guennouni 2004) In that sense, the SARL can be understood as “a hybrid form which in parts draws its rules from joint stock companies and in parts from partnerships. It is an ideal form for the operation of small- and medium-sized enterprises.”10 (Msalha 2005, 109) This distinction — between a large formal enterprise comprising many contributors (the soci´et´e anonyme, literally the ‘anonymous company’) and smaller, more informal firms — underpins the many distinctions in substantive law. These differences can be understood along a number of dimensions.
Capital: In line with French law, both the SA and SARL forms have a minimum capital requirement. However, that requirement is much higher in the case of the SA; an SARL must have listed capital of at least 100 000 dirhams,11 while an SA must have a minimum capital of 300 000 dirhams, or 3 million dirhams if the company issues a prospectus.12 Similarly, there are significant differences in the transferability of capital: shares in an SA are freely transferrable (Martin 1999, 163), whereas shares 9
10
11 12
Original quote: “. . . la soci´et´e anonyme a vocation a` eˆ tre une soci´et´e de capitaux et un instrument de drainage de l’´epargne publique. Elle se distingue ainsi tr`es nettement de la SARL, structure destin´ee aux entreprises de taille moyenne et/ou a` caract`ere familial, o`u c’est la personnalit´e de l’apporteur qui compte en principe davantage que le capital apport´e.” Original quote: “. . . la soci´et´e a` responsabilit´e limit´ee s’apparente a` la fois aux soci´et´es de capitaux et aux soci´et´es de personnes. C’est un type hybride qui puise ses r`egles tantˆot des soci´et´es de capitaux, tantˆot des soci´et´es de personnes. C’est un outil id´eal pour l’exploitation des petites et moyennes enterprises.” Article 46, law 05-06. Article 6, law 17-95.
8. Appendix: Details of the legal reform
iii
in an SARL generally may not be transferred to third parties without the consent of a majority of shareholders representing three-quarters of the shares (Martin 1999, 157158).13 Finally, the SA has much wider scope for dealing in its capital: unlike the SA, the SARL may not grant security over moveable interests,14 nor represent any of its listed capital by negotiable instruments.15
Governance: The SA and SARL forms exhibit clearly different governance structures. An SA may choose between two governance structures, both dualist and both quite onerous: a Board of Directors with an Administrator or a Board of Directors overseen by a Board of Trustees.16 The significance of this will be discussed in comparing the new regime with the old; for now, note merely that both structures provide significant division of management powers. In contrast, an SARL has a unitary structure: it is managed merely by one or more designated managers,17 empowered (in the absence of specific limitation in the company statutes) to take all necessary action in the interests of the company.18
Transparency: Improvements in corporate transparency are an important part of the legal reforms, as subsequent discussion will show. However, understandably, there remain significant differences in transparency standards between the SA and SARL form under the new law. An SA must have at least one designated auditor at all times (and at least two, if it issues a prospectus).19 In contrast, while an SARL may appoint an auditor, appointment is obligatory only in two cases: (i) where the company has a sales turnover of at least 50 million dirhams (net of taxes), and (ii) where the appointment is granted by a judge on the application of one or more shareholders rep13 14 15 16 17 18 19
Articles 58 and 60, law 05-96. Article 54, law 17-95. Article 55, law 17-95. See Chapter 1 and Chapter 2 respectively, Titre 3, law 17-95. Article 62, law 05-96. Article 63, law 05-96. Article 159, law 17-95.
8. Appendix: Details of the legal reform
iv
resenting at least a quarter of the listed capital (Martin 1999, 162).20 (In the event that an SARL does appoint an auditor, the auditor has the same powers, obligations and responsibilities as in the SA case.21 )
8.3
The old and the new: a summary of the legal reform
Equally fundamental to the subsequent analysis is the recognition that the legal reform indeed reflected a significant change; in particular, that the reform significantly increased the relative cost of retaining the SA status. That is the key point that the following summary seeks to make.
It is difficult, as a matter of research, to identify precisely the changes wrought by the reforms. Unlike the new SA and SARL statutes (which can be purchased even from street vendors in Casablanca), the text of the old law is not readily available (and indeed has not been directly consulted for this summary). Similarly, and perhaps somewhat incongruously, academic literature on the reform (such as that cited in this summary) provides little detail on the previous legal regime. Thus, this summary draws primarily upon general discussions of the reform provided in L’Economiste, the leading Moroccan business newspaper. For that reason, the following summary must essentially be limited to outlining the reforms regarding the SA form. However, even in doing so, it is clear that this form attracted by far the more significant and more controversial overhaul; while the reform process did involve a new statute for the SARL form, this law “has not raised much interest”22 (El Hammoumi 2005), as even a casual survey of L’Economiste articles reveals. In effect, it appears that the law on SARL firms, while reformulated, has remained largely unchanged; in contrast, the SA law has undergone a radical overhaul, so that the SA form now resembles the SARL form much less than 20 21 22
Article 80, law 05-96. Articles 13 and 83, law 05-96. Original quote: “La loi n 05-96, relative aux autres formes de soci´et´es commerciales, n’a pas soulev´e autant d’int´erˆet.”
8. Appendix: Details of the legal reform
v
it previously did. This section outlines the key areas of change.
Governance: The previous law reflected a disturbing dichotomy: by providing little guidance on the role of corporate directors, the law allowed both for some directors to have nearly unchecked power while other directors were allowed to play almost no part in the running of the company. Thus, in advocating the reform, the Minister for Privatisation (Abderrahmane Saa¨ıdi) argued, “We do not know who does what. The president is often invested a function of prestige without assuming any real responsibilities.”23 (Mossadaq and Oudghiri 1995) Concomitantly, many expressed concern that “[a]dministrators were reduced by the old law to ratifying decisions taken by the ‘all-powerful’ president”24 with the only real limitation upon the president’s power being the corporate object itself (Shamamba 1995).25 Law 17-95 brought radical change in this area. As explained earlier, the law introduces a ‘dualist’ system of governance, “inspired by German law, and well known to French corporate law”,26 involving either a Board of Directors and President or a Board of Directors and Board of Trustees; in both cases, providing new checks upon management authority (Oudghiri 1998). In doing so, the law seeks to enliven so-called ‘sleeping directors’ to be more involved in corporate management; it “encourages creation within the Board of Directors of technical committees charged to study particular questions and to formulate opinions and recommendations.”27 (Alami 1999)
Transparency: A significant complaint under the old regime was the untrustworthiness of corporate accounts. Moroccan firms, it was often claimed, would routinely 23 24
25 26 27
Original quote: “Nous ne savons pas qui fait quoi. Le pr´esident est souvent investi d’une fonction de prestige sans pour autant assumer de v´eritables responsabilit´es”. Original quote: “Les administrateurs en e´ taient r´eduits a` ent´eriner des d´ecisions prises par le “tout puissant” pr´esident.” The article was paraphrasing comments by Abdelaziz Squalli, Professor at the Facult´e de Droit de F`es. Original quote: “A ce jour, la seule limite connue au pouvoir du PDG e´ tait l’objet social.” Original quote: “Notion nouvelle inspir´ee du droit allemand et bien connue du droit franc¸ais des affaires” Original quote: “En effet, l’article 51 encourage la cr´eation au sein du conseil d’administration de comit´es techniques charg´es d’´etudier des questions pr´ecises et de formuler des avis et recommandations.” See Article 51, Law 17-95.
8. Appendix: Details of the legal reform
vi
produce three sets of accounts: one for the manager, one for the tax authorities and one for the bank.28 Law 17-95 is one important component of reforms to counter this problem.29 The new conditions regarding auditors have already been outlined; it bears noting that, under the previous regime, auditors did not have a ‘permanent presence’ in the company’s operations (Oudghiri 1998).30 Thus, the reform also strengthens the effect of the minimum capital requirement; it “puts an end to practices of posting a fictitious capital having no relationship with the capital actually contributed.”31 (L’Economiste 1998b)
This relates closely to new rights to information: unlike the previous regime, law 1795 “envisages the shareholder having access to all documents necessary to understand the management of the company. It also involves obligations to provide information in the management report relevant to subsidiary companies, significant shareholdings and takeovers.”32 (Oudghiri 1998) Finally, the transparency requirements bring new standards regarding conflicts of interest. “In effect, under the former law, a director was able to make contracts with the company, to borrow money from the company and to guarantee his or her own debts by the company, all without the shareholders being informed of these actions.”33 (Alami 1999) These practices are ended by the new statute, which requires such contracts to be subject to prior authorisation by the Board of Directors, to be the subject of an auditor’s report and to be approved by the 28 29
30 31 32
33
One businessman interviewed for this research wryly suggested a fourth set: for the manager’s wife. There are other components here, too; for example, an initiative of the Central Bank of Morocco (Bank Al-Maghrib) that requires every credit file to have attached a copy of the financial statements that were filed with the Commerce Tribunal. Original quote: “. . . la pr´esence permanente des commissaires aux comptes. . . ”. Original quote: “Cette disposition met ainsi fin aux pratiques qui consistaient a` afficher un capital fictif sans rapport avec celui r´eellement apport´e.” Original quote: “La loi pr´evoit la mise a` la disposition des actionnaires de tout document n´ecessaire a` l’appr´eciation de la gestion de la soci´et´e. Il est e´ galement fait obligation de mentionner les informations relatives aux filiales, participations et prises de contrˆole dans le rapport de gestion.” See Titre V, law 17-95. Original quote: “En effet, sous l’emprise de l’ancienne loi, un administrateur pouvait passer des contrats avec la soci´et´e, emprunter de l’argent a` la soci´et´e, faire cautionner par la soci´et´e ses engagements personnels, sans mˆeme que les actionnaires soient inform´es de ces op´erations.”
8. Appendix: Details of the legal reform
vii
next ordinary general meeting.34
Minority rights:
The previous law did not grant genuine rights to minority share-
holders; thus, it was said of those SA companies not bringing their statutes into early compliance with the new law that “they may continue to exercise on minorities the dictatorship of the majority”35 (Oudghiri and Ikram 1999) The new law redresses this. Aside from the right to information and the expanded role for auditors just outlined, the legislation allows individuals or groups of shareholders holding ten percent of the capital to demand the convening of a general meeting.36
Third parties: Under the old law, third parties had little legal reassurance in dealing with a company; because the relations between companies and their administrators were merely contractual, a third party always bore the risk that a company administrator having ostensible authority was not, actually, authorised to deal on the company’s behalf. The new law rectifies this, by implementing what English law knows as the ‘indoor management rule’: “[a]ny limitation of the powers of the capacities of administrators may not be constested as against third parties. Third parties may, consequently, act with reassurance: the law protects them.”37 (Alami 1999) Further, whereas the old law allowed the company to take legal personality from the time of the constitutive meeting, the new law requires registration on the Register of Commerce as a precondition; this provides reassurance to third parties, particularly those needing to deal with ‘founders’ of the company prior to registration, by making the corporate status of the company a matter of public record (Oudghiri 1998).
34 35 36 37
Articles 56–62, Law 17-95. Original quote: “Ils pourront continuer exercer sur les minoritaires la dictature des majoritaires.” Article 116, law 17-95. Original quote: “Toute limitation des pouvoirs des administrateurs est inopposable aux tiers. Ces derniers peuvent par cons´equent agir en toute qui´etude: la loi les prot`ege.”
![[PDF BOOK] Evaluating the Impact of Leadership Development: A ...](https://p.pdfkul.com/img/300x300/pdf-book-evaluating-the-impact-of-leadership-devel_59bb23811723dde3a979bfdc.jpg)
