How does bank credit affect the shape of business groups' internal capital markets?

ABSTRACT

In empirical literature, it is hypothesized that the persistence of Business Groups (BGs) is linked to the capability of easing the financing constraints of participating firms through the implementation of an internal capital market (ICM). ICMs enable capital relocation, thus partially offsetting disparities in accessing bank credit. I offer three contributions. First, I formally ground this idea with a dynamic model in which an ICM is endogenously generated as a function of the bank's lending policy and firms' production incentives. A novelty in the literature, the model contains a trickle-down mechanism which allows bank credit to circulate across firms via inter-firm loans. Second, I study the model to understand how the ICM reacts to shocks to the following empirically critical channels: the BG's debt-to-equity ratio, the profitability of production markets and, most importantly, the bank's credit rationing policy. The model disentangles the contribution of each channel on the shape of the ICM, as measured in terms of the intensity of firms' cross-subsidization. In particular, I discover a non-monotonic relationship between the latter dimension and the intensity of cross-subsidization. Third, I match stylized facts of the so called ‘Korean crisis’ and apply the model to discipline some extant results of the empirical literature: I find that a tightening of the financial market, taken in isolation, would contribute to a densification of the ICM structure, reflecting the functioning of the credit-reallocation channel described in the literature. At the same time, the magnitude of such effect is not sufficiently strong to overcome the reduction in density as caused by the two alternative channels.

Keywords:

• Internal capital markets
• Credit-crunch
• Financial networks
• Business groups

JEL Classifications:

• G31
• G32
• G34

Acknowledgements

The author is particularly indebted to the editor J. Gatheral, an anonymous referee, S. Currarini, T. Gall and A. Ianni for extremely helpful suggestions. I also warmly thank M. Anufriev, P. Bertolini, P. Campana, S. Corsi, S. Della Lena, D. Goldbaum, C. Di Guilmi, S. Galanis, A. Laiho, A. Landi, A. Manzini, G. Marotta, A. Mennuni, F. Nava, V. Panchenko, P. Pin, A. Rosato, the participants of AFBC2017, the Asian Meetings of Econometric Society 2017 and attendees to seminars at Southampton, Modena and Paris 1 for useful comments. This research was supported by ESRC, Fondazione Banca Popolare di Novara and the Australian Research Council through Discovery Project (DP170100429). The author declares that he has no relevant or material financial interests that relate to the research described in this paper.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Gopalan et al. (2007) provide evidence that more than 70% of intra-group loans in Indian business groups are funded by external debt financing. I am grateful to a referee for pointing this out.

2 In the model, I capture the high degree of idiosyncrasy governing inter-firm capital flow transactions characterizing empirical ICMs (see, for example, Khanna and Yafeh 2007) by assuming that exchanges are subject to pairwise bargaining between PUs.

3 See, for a broad discussion on the causes and the consequences of the Korean crisis, Almeida et al. 2015 and Lee et al. 2009.

4 Empirically, this coincided with a regulatory shock that prohibited cross-loan guarantees and cross-shareholdings forced debt-to-equity ratios under 200% (Lee et al. 2009).

5 See Jackson and Rogers (2007), Vega-Redondo (2007) for a discussion of recent contributions. Random networks allow to project the effect of local interactions into the structural properties of a resulting aggregate system. However, with respect to the cited models, the formation mechanism here is pairwise efficient and is driven by micro-founded incentives.

6 In a seminal contribution, Boot et al. (1993) underlines the ubiquity of state-contingent guarantees between members of holding-groups (e.g. comfort letters, that is promises of rescue sent by the parent to the affiliates lenders) and provides a reputation-based mechanism for which conditional guarantees may be preferred to enforceable cross-guarantees as they provide the holding company with flexibility in managing financial impairment together with the benefits of legal separation among the holding company entities. I refer the reader to the literature review of Luciano and Nicodano (2014) for evidence on the use of intra-group guarantees in empirical BGs.

7 I am profoundly grateful to an anonymous referee who provided excellent guidance for restructuring this and the next sections.

8 See for instance (Almeida et al. 2011, 2015) for a discussion of the real effect of regulations on the empirical counter-part of α and β in the context of Korean Chaebols.

9 Technically, this assumption enables the unique stationary distribution of inter-firm loans (see section 3.1).

10 Note, however, that the Assumption is moderated by the fact that in this paper intra-group exchanges have to be incentive-compatible on pairwise basis to realize (see section 2.6).

11 Since corporate regulations α and β dictating inter-firm exchanges are unknown to the bank, the bank will take firms' sales $p\cdot q_i$ as random variable.

12 The introduction of an intermediate case with heterogeneous cash endowments would allow to consider a scenario in which loans descending from bank credit can be (partially) crowed-out in favor of intra-group credit exchanged based upon internal resources. Albeit heterogeneous cash endowments can be a relevant venue for studying certain phenomena (such as the resilience of business groups to external shocks Allen et al. 2019), it would not alter the effects of bank discrimination on the topology of the ICM, the focus of the present study.

13 The use of entrusted loans for inter-firm capital flows instead of governance-related mechanisms (e.g. share issuance) and the related wide spectrum of interest rates is predominant in many types of ICMs (Allen et al. 2019; Buchuk et al. 2014). The bargaining mechanism and the pairing device I devise are agnostic on the degree of control exerted by lenders on borrowers, thus allowing for a broad spectrum of (noisy) configurations. Furthermore, the resulting formation protocol is such that economic incentives, rather than property rights are the main motivating driver for the formation of ICMs.

14 The assumption simplify non-essential comparative statics and can be easily relaxed with no effect on any of the main results.

15 By imposing c = r it is immediate to recover from Assumption 2.1 the spread structure as described in the empirical literature (see for instance Allen et al. 2019).

16 Looking at aggregate leverage ratios, Buchuk et al. (2014) find that recipients of intra-group transfers in Chilean business groups have leverage ratios that are $7-10\mathrm{\%}$ higher than non-conglomerated firms.

17 As the angle of this paper is to isolate the contribution of a trickle-down financing mechanism to the formation of ICMs, I deviate from the traditional contract-theoretic framework (such as Gertner et al. 1994) and condense all contract-theoretic concerns in the bank-firm relationship, thus assuming perfect enforcement at the firm-firm relationship.

18 The rationing proposed here is consistent with the empirical finding of a large literature investigating the nexus between bank rationing and BG performance (see, for example, Gopalan et al. 2007).

19 Banks and regulators may enforce a tight interpretation of loss given default over a single position. For instance, by netting the collateral from a range of standard expected penalties (such as litigation costs) that become apparent once the project fails, the bank may estimate a χ within the more eventful interval $[-1,1]$. By analyzing a large sample of small business non-performing bank loans, Eales and Bosworth (1998) provide strong evidence of a convex relation between severity and frequency of loss given default. Relevantly, losses given default exceeding by up to the $20\mathrm{\%}$ of the principal (i.e. $\chi =-0.2$) are reported twice the frequency with respect to cases with recovery rates range in $\chi \in [0.7,0.9]$.

20 This simple risk structure instruments the potential firm's operational risk (see Freimer and Gordon 1965) with a bound on the firm's return on asset (ROA). In the model, operations are simply given by input transformation and marketing of output.

21 Given two univariate distributions $p_{\omega }$ and $p_{\omega }^{\mathrm{\prime }}$, $p_{\omega }$ First-Order Stochastic Dominates $p_{\omega }^{\mathrm{\prime }}$ or, equivalently, $p_{\omega }\stackrel{\mathrm{F}\mathrm{O}}{\succ }{p}_{\omega }^{\mathrm{\prime }}$ if $p_{\omega }\le p_{\omega }^{\mathrm{\prime }}\mathrm{\forall }\omega \in \mathrm{\Omega }$. This is to say that the mass of the distribution moves upward on the support when the distribution shifts from $p_{\omega }^{\mathrm{\prime }}$ to $p_{\omega }$.

22 Notice that it is such incentive structure that opens up an interesting arbitrage opportunity where borrower $t\in N_t$ becomes a potential lender for borrower $t^{\mathrm{\prime }\mathrm{\prime }}>t$ accessing the ICM in next periods, thus effectively spreading the bank's capital through the ICM, coherently with a pervasive empirical regularity of ICMs (see, for example, Almeida et al. 2011, Buchuk et al. 2014), where capital trickles-down from larger and more mature low-productivity firms to younger and smaller high-productivity firms.

23 Gopalan et al. (2007) show that interest rates charged by lenders can have only a latent connection with those of formal money market in Indian Business groups. By looking at Chinese entrusted loans, official interest rates act as a further lower bound (Allen et al. 2019). A mechanism based upon exchange of liquidity against internal equity would prioritize strategic incentives rather than financing incentives behind the ICM. It is unclear which motif is predominant (see Buchuk et al. (2014) for a discussion). Furthermore, the second mechanism would put a strong assumption on the direction of capital flows.

24 Importantly, the evolution of $s_{t^{\mathrm{\prime }}}\left(t\right)\ge 0$ will proxy the analysis of ICM formation in section 3.

25 Furthermore, in section 4.1, I confirm in simulations the validity of the assumptions made in this section.

26 Furthermore, out of the present model where external finance is placed ex-ante the ICM formation, the approximation in (17(17) ${\overline{\iota }}_{K_{t^{\mathrm{\prime }}}\left(t\right)}^l\approx m\left({\omega }_{t^{\mathrm{\prime }}}\right)\cdot s_{t^{\mathrm{\prime }}}(t-1)\cdot \ell ,$(17) ) appears to be reasonable under the empirical observation that firms can go through several rounds of external financing (Khanna and Yafeh 2007). Hence, a model with dynamical bank re-financing would allow to organically incorporate the assumption.

27 Hence, the distribution-based approach avoids an analysis based upon ad hoc ICMs structures, and as such can be tested against data.

28 Differently from other works (see Jackson and Rogers 2007) in which FOSD is assessed for a uni-variate network distribution $P\left(s\right)$, I show that FOSD does hold also in a bivariate case such the one I presented with $P(s,\omega )$. Given two bivariate distribution $P(s,\omega )$ and $P^{\mathrm{\prime }}(s,\omega )$, I say that $P(s,\omega )$ first order stochastically dominates $P^{\mathrm{\prime }}(s,\omega )$ if $P(s,\omega )\le P^{\mathrm{\prime }}(s,\omega )\mathrm{\forall }s>0,\omega \in \mathrm{\Omega }$.

29 I refer the reader to the literature in Khanna and Palepu (2000), Khanna and Tice (2001), Buchuk et al. (2014) for a comprehensive set of examples.

30 I thank an anonymous referee for critical suggestions that led to the results of this section.

31 The full taxonomy is analyzed in the proof of Proposition 3.6.

32 For example, in discussing the effects on Korean firms of interest rate hikes during the Korean Crisis, Borensztein and Lee (2002) find that real money and credit slowed down during the crisis, yet, it is not clear on how significantly domestic credit conditions tightened, as both credit supply and demand felt simultaneously.

33 More precisely, the bank lending policy ${\omega }_e$ pivots around the pivotal type ${\hat{e}}_{p,r}$.

34 For a comprehensive description of the crisis, I direct the reader to Almeida et al. (2015) and the literature therein cited.

35 It is indeed possible to construct a more refined endowment distribution using, for example, the Korea Listed Companies Association Data-set (see, for instance, Lim 2012).

36 In untabled simulations I try alternative starting values with no effect on the qualitative results presented here.

37 Hence, I adopt a conditional FOSD, which is a stronger ordering with respect to FOSD. Indeed it is not generally the case that FOSD is preserved under truncation (see for instance Riley 2012, p. 249).

38 This implies a non-negative correlation between ω and s. Without such restriction, conditional dominance may not imply joint dominance (see Levy and Paroush 1974, p. 615).

39 In untabled results, I show that the introduction of $1/\overline{e}{B}_{r,\chi }\left(\tau \right)$ makes the computation of (A9(A9) $\begin{array}{rl}{\overline{\psi }}_{r,\chi ,\alpha ,{\omega }^{\ast \ast }}\left(\tau \right)& \equiv \sum _{e=0}^{e_1\left(\tau \right)}p\log (1+\alpha )\\ & -\sum _{e=e_1\left(\tau \right)+1}^{e_2\left(\tau \right)}p\log \left(A_{p,r}\right(\tau )+e{B}_{r,\chi }(\tau \left)\right).\end{array}$(A9) ) more cumbersome without affecting the qualitative result of the proof.

40 The sum notation is simply used to emphasize the effect of changes to the types' structure as induced by changes to $e_1\left(\tau \right)$ and $e_2\left(\tau \right)$ but it does not alter the qualitative structure and result of the proof.

Additional information

Funding

This research was supported by ESRC, Fondazione Banca Popolare di Novara and the Australian Research Council through Discovery Project [DP170100429].