Financial Stability and Interacting Networks of Financial Institutions and Market Infrastructures

Transcription

1 Financial Stability and Interacting Networks of Financial Institutions and Market Infrastructures Seminar on Network Analysis and Financial Stability Issues Mexico City, Mexico, December 10 and 11, 2014 Carlos León Banco de la República & Tilburg University Ron J. Berndsen De Nederlandsche Bank & Tilburg University Luc Renneboog Tilburg University & European Governance Institute

2 Based on León C., Berndsen R.J., & Renneboog, L., Financial Stability and Interacting Networks of Financial Institutions and Market Infrastructures, Borradores de Economía, No.848, Banco de la República, Disclaimer: The opinions and statements in this article are the sole responsibility of the authors and do not represent neither those of Banco de la República nor of De Nederlandsche Bank. Comments and suggestions from Clara Lía Machado, Joaquín Bernal, Freddy Cepeda and Jhonatan Pérez are appreciated. Helpful assistance in data processing and visualization from Carlos Cadena and Santiago Hernandez is greatly appreciated. Any remaining errors are our own.

3 Five take-home ideas 1. Colombian financial institutions networks display a modular scale-free architecture, consistent with previous evidence of inhomogeneity and hierarchical organization. 2. If several financial institutions networks are coupled, the resulting multiplex network preserves the modular scale-free architecture because of positively correlated multiplexity. 3. However, if financial market infrastructures (FMIs) are considered as the plumbing providers for financial institutions (FIs) in an interacting network, modularity vanishes. 4. As modularity is a source of resilience (i.e. isolate feedbacks and limit cascades), authorities pursuing financial stability should examine the financial multiplex and interacting networks. 5. The role of financial market infrastructures in financial stability should not be underestimated (i.e. infrastructure-related systemic risk)

4 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

5 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

6 The modular scale-free architecture Extensive evidence on financial networks inhomogeneity: Boss et al. 2004; Soramäki et al., 2007; Inaoka et al., 2004; Bech & Atalay, 2010; Bargigli et al., 2013; Craig & von Peter, 2014; Martínez-Jaramillo et al., 2012; in t Veld & van Lelyveld, Scale-free networks (Barabási & Albert, 1999) are the customary model in social sciences; some financial networks fit too. But, scale-free networks are not hierarchical in nature (Ravasz & Barabási, 2003) Evidence of tiering (i.e. hierarchies) in financial networks (as in Craig & von Peter, 2014) contradicts scale-free networks. Core-periphery models may fit financial networks better than the scalefree model (as in Craig & von Peter, 2014).

7 The modular scale-free architecture What about modular scale-free networks (Dorogovtsev, 2002; Barabási, 2003; Assenza et al., 2011)? Scale-free networks A few well-connected throughout the entire network. Non-hierarchical in nature (Ravasz & Barabási, 2003) Robust yet fragile (Haldane, 2009) Modularity Sparsely interconnected neighborhoods of dense interaction. Hierarchy: system of subsystems (Simon, 1962) Change can be isolated to local neighborhoods (Anderson, 1999)

8 The modular scale-free architecture What about modular scale-free networks (Dorogovtsev, 2002; Barabási, 2003; Assenza et al., 2011)? Scale-free networks A few well-connected throughout the entire network. Non-hierarchical in nature (Ravasz & Barabási, 2003) Robust yet fragile (Haldane, 2009) Modular scale-free A few-well connected lead densely connected clusters, and serve as inter-cluster conduits. The well-connected are the source of fragility, but they serve as firewalls or circuit breakers against contagion. Robust and resilient, yet fragile. Modularity Sparsely interconnected neighborhoods of dense interaction. Hierarchy: system of subsystems (Simon, 1962) Change can be isolated to local neighborhoods (Anderson, 1999)

9 The modular scale-free architecture What about modular scale-free networks (Dorogovtsev, 2002; Barabási, 2003; Assenza et al., 2011)? Most salient feature (detection): Inverse relation between local clustering and degree In the Colombian case, evidence supports the modular scale-free architecture of financial networks (see León & Berndsen, 2014)

10 Generation Implications Description The modular scale-free architecture About modularity: Newman (2003): high intra-group density, low inter-group density. Simon (1962): clusters of dense interaction or nearly decomposable systems. Battiston et al. (2009): agents clustered in neighborhoods, with connections among neighborhoods being sparse. Anderson (1999): subsystems receive inputs from few other components, thus change can be isolated to local neighborhoods. Haldane & May (2011): by limiting potential cascades, modularity protects the systemic resilience of both natural and constructed networks. Assenza et al. (2011): growth + preferential attachment + homeostasis (i.e. intensity preservation mechanism). Battiston et al (2009): costs of assessing credit worthiness and maintaining relationships yield sparsely connected neighborhoods (homeostasis?)

11 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

12 Financial multi-layer networks Networks of networks Multiplex: Participants of one sort but several kinds of edges (Baxter et al., 2014) Different layers are modeled by means of different types of links (D Agostino et al., 2014) Interacting: Multi-layer networks of distinct types of participants that relate across networks Different layers are explicitly modeled as separate networks and the links among them represent the inter-layer interactions (D Agostino et al., 2014)

13 Financial multi-layer networks Example: A two-layer multiplex of FIs buying/selling FX and making large-value payments Large-value payments FX Multiplex settlement network How are this transactions settled in real-life? Figure 1. A multiplex network. Two-layer networks, X and Y, and the multiplex (Z) resulting from merging X and Y. Vertical lines connecting superimposed vertexes are the participants, whereas each vertex is a role in the corresponding layer.

14 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX (USD) ABC XYZ

15 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX (USD) ABC XYZ Hint: settlement of transactions between FIs is not bilateral!

16 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX (USD) ABC FX settlement FMI XYZ Financial Market Infrastructures (FMIs): multilateral systems providing trading, clearing, settling, recording, and compressing services for transactions between FIs. (We focus on FMIs providing settlement services) FMIs are the plumbing of the financial system (Bernanke, 2011). In the absence of FMIs, FIs would not be able to settle most of their transactions.

17 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX

18 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX

19 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX

20 Financial multi-layer networks How do financial transactions settle in real-life? Say, buying FX If several FIs are considered

21 Financial multi-layer networks How do FMIs interact with the two-layer multiplex? FIs FMIs FX Settlement system Large-value payment system

22 Financial multi-layer networks How do FMIs interact with the two-layer multiplex? FIs FMIs In the absence of the link between FIs and FMIs, transactions could not be settled (i.e. dependence links, Gao et al., 2012) In the absence of the link among FMIs, transactions could not be settled FMIs network is a critical infrastructure for the financial system

23 Financial multi-layer networks Any other cases outside the financial realm?

24 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

25 The datasets From three FMIs, information to build three settlement networks in the Colombian financial market (2012): Other Sovereign securities market Foreign exchange market FX Sovereign Securities Other markets (equity, derivatives, interbank funds, corporate) Daily data for Results will be presented as daily averages. The FIs multiplex network is built by aggregating the three single-layer settlement networks. Afterwards, the FIs & FMIs interacting network is build by considering the role of each FMI in the settlement of each individual transacion

26 The datasets The Colombian settlement multiplex* Sovereign Securities FX Other Each vertex corresponds to a FI potentially fulfilling multiple roles in the Colombian financial system, whereas each arrow and its width represent the existence of a payment between FIs and its monetary value, respectively. (*) Multiplex: a network containing participants of one sort (i.e. FIs) but several kinds of edges (i.e. sovereign securities, foreign exchange, interbank, etc.). Based on Baxter et al. (2014).

27 (*) Each vertex in the first (upper) layer corresponds to a FI potentially fulfilling multiple roles in the Colombian financial system, whereas each vertex in the second layer corresponds to a FMI. The diameter of all vertexes is determined by their strength (i.e. the value of the ingoing and outgoing weighted connections). Each arrow corresponds to a part of the settlement process. The datasets The Colombian settlement interacting network* FX settlement system (CCDC) Large-value payment system (CUD) Sov. Securities settlement system (DCV)

28 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

29 Statistic Sovereign securities market Table 1 Basic statistics of the networks a Monoplex networks (in Figure 3) Foreign exchange market Other markets n d μ k Multiplex network (in Figure 4) σ kin /out 8.60/ / / /13.44 γ kin /out 3.12/ / / /2.49 γ sin /out 1.77/ / / /1.99 l 2.24 [~4.9] 1.83 [~3.8] 2.34 [~5.0] 2.19 [~5.0] c 0.15 [~0.0] 0.24 [~0.0] 0.11 [~0.0] 0.17 [~0.0] c w 0.17 [~0.0] 0.28 [~0.0] 0.20 [~0.0] 0.25 [~0.0] r kin /out 0.31/0.32 [~0.0] 0.59/0.60 [~0.0] 0.23/0.22 [~0.0] 0.38/0.36 [~0.0] r sin /out 0.19/0.20 [~0.0] 0.34/0.37 [~0.0] 0.12/0.09 [~0.0] 0.15/0.13 [~0.0] γ ci This table shows that the basic statistics of the monoplex networks and the resulting multiplex approximate to those of a modular scale-free network. a Statistics presented are: number of vertexes (n); density (d); average degree (μ k ); in/out degree standard deviation (σ kin /out ); in/out degree Power-law exponent (γ kin /out ); in/out strength Power-law exponent (γ sin /out ); mean geodesic distance (l); clustering coefficient (c); degree correlation (r kin /out ); strength correlation (r sin /out ); local clustering power-law exponent (γ ci ). Expected values for large random networks are reported in brackets. Sparse networks Poorly connected Consistent with related literature But 2-3 degrees of separation!! Ultra-small (Cohen & Havlin, 2010) Spreading is very efficient Consistent with related literature And Clustered Tendency to share counterparties Consistent with related literature See also León & Berndsen (2014)

30 Statistic Sovereign securities market Table 1 Basic statistics of the networks a Monoplex networks (in Figure 3) Foreign exchange market Other markets n d μ k Multiplex network (in Figure 4) σ kin /out 8.60/ / / /13.44 Scale-free networks By links (degree) By values (strength) γ kin /out 3.12/ / / /2.49 γ sin /out 1.77/ / / /1.99 l 2.24 [~4.9] 1.83 [~3.8] 2.34 [~5.0] 2.19 [~5.0] c 0.15 [~0.0] 0.24 [~0.0] 0.11 [~0.0] 0.17 [~0.0] c w 0.17 [~0.0] 0.28 [~0.0] 0.20 [~0.0] 0.25 [~0.0] r kin /out 0.31/0.32 [~0.0] 0.59/0.60 [~0.0] 0.23/0.22 [~0.0] 0.38/0.36 [~0.0] r sin /out 0.19/0.20 [~0.0] 0.34/0.37 [~0.0] 0.12/0.09 [~0.0] 0.15/0.13 [~0.0] γ ci This table shows that the basic statistics of the monoplex networks and the resulting multiplex approximate to those of a modular scale-free network. a Statistics presented are: number of vertexes (n); density (d); average degree (μ k ); in/out degree standard deviation (σ kin /out ); in/out degree Power-law exponent (γ kin /out ); in/out strength Power-law exponent (γ sin /out ); mean geodesic distance (l); clustering coefficient (c); degree correlation (r kin /out ); strength correlation (r sin /out ); local clustering power-law exponent (γ ci ). Expected values for large random networks are reported in brackets. Assortative network High (low) degree vertexes tend to be interconnected Also by strength And Poorly-connected are incestuous Well-connected are not A modular hierarchy exists according to Dorogovtset et al. (2002) and Barabasi (2003). See also León & Berndsen (2014)

31 Results The multiplex preserves the main properties of the monoplex networks Positively correlated multiplexity: a vertex with large degree in one layer likely has more links in the other layer as well (Lee et al., 2014; Kennet et al., 2014) k sov k fx k rest s sov s fx s rest k sov 1 k fx k rest s sov 1 s fx s rest Figure 6. Monoplex networks degree and strength correlation matrix. The correlation matrix was estimated based on the contribution of each FI to the total degree or strength in the whole samples.

32 Results The multiplex preserves the main properties of the monoplex networks Positively correlated multiplexity: a vertex with large degree in one layer likely has more links in the other layer as well (Lee et al., 2014; Kennet et al., 2014) By degree By strength Figure 7. Positively correlated multiplexity. Participating FIs in each layer are ranked in decreasing order of degree (left) and strength (right) in the corresponding horizontal axis. High-degree and high-strength FIs in one layer tend to be the high-degree in the other two layers, which confirms the presence of positively correlated multiplexity.

33 Statistic Table 2 Basic statistics of the networks a Multiplex Network (in Figure 4) n d μ k Interacting Networks (in Figure 5) σ kin /out 13.41/ /9.75 γ kin /out 2.81/ /3.36 γ sin /out 1.96/ /1.77 l 2.19 [~5.0] 1.99 [~5.0] c 0.17 [~0.0] 0.01 [~0.0] c w 0.25 [~0.0] 0.14 [~0.0] r kin /out 0.38/0.36 [~0.0] -0.43/-0.17 [~0.0] r sin /out 0.15/0.13 [~0.0] -0.31/-0.15 [~0.0] γ ci This table shows that the basic statistics of the multiplex network approximate to those of a modular scale-free network, whereas the interacting network s to those of a scale-free network only. a Statistics presented are: number of vertexes (n); density (d); average degree (μ k ); in/out degree standard deviation (σ kin /out ); in/out degree Powerlaw exponent (γ kin /out ); in/out strength Power-law exponent (γ sin /out ); mean geodesic distance (l); clustering Sparse networks Poorly connected Approximate scale-free networks By links (degree) By values (strength) 2 degrees of separation Ultra-small (Cohen & Havlin, 2010) Spreading is very efficient Clustering (by degree) vanishes Disassortative network High (low) degree vertexes tend to be connected to low (high) degree vertexes Also by strength Modularity vanishes

34 Modularity vanishes Figure 8. Distribution of local clustering as a function of average degree. Bas entire data sample, there is evidence of an inverse relation between average d clustering for the FIs multiplex, whereas such relation is absent in the FIs a interacting network. Figure 8. Distribution of local clustering as a function of average degree. Based on the entire data sample, there is evidence of an inverse relation between average degree and clustering for the FIs multiplex, whereas such relation is absent in the FIs and FMIs interacting network.

35 Contents The modular scale-free architecture Financial multi-layer networks The datasets Main results The critical role of FMIs for financial stability

36 The critical role of FMIs for financial stability In the Colombian monoplex and multiplex networks there is evidence of a modular scale-free architecture: robust and resilient, yet fragile networks (see León & Berndsen, 2014) Ignoring the connective role of FMIs within financial networks may mislead the analysis of the architecture of financial systems. The main consequence of coupling FIs and FMIs networks is the removal of modular hierarchy, which invalidates the presumption of a financial architecture that favors systemic resilience. In the absence of modularity, there are no subsystems of FIs, and they tend to receive inputs from all other FIs via FMIs, thus changes are not isolated and tend to spread across markets and their participants.

37 The critical role of FMIs for financial stability An economic rationale: The purpose of settlement FMIs: the centralized extinction of claims between FIs. FMIs provide an alternative to the frictions that arise when money and financial securities are traded directly (Manning et al., 2009). As their purpose is the efficient and safe flow of money and financial securities to all FIs, it is not surprising that FMIs remove modularity and may act as conduits for widespread contagion.

38 The critical role of FMIs for financial stability All in all The well-functioning of the FMIs network determines the extent to which the benefits of the modular scale-free architecture of FIs networks apply. The well-functioning of the FMIs network determines whether the settlement of financial transactions is carried out or not. The interdependencies between FIs and FMIs networks make the financial system more fragile: damage to one FMI (e.g. operational or financial) can trigger a catastrophic cascade of events that propagates across the global connectivity. (CCPs?) Pending issues Additional layers of complexity are readily available to expand our interacting network of FIs and FMIs (physical infrastructures: e.g. communications, power) Linkages with FIs and FMIs abroad? (e.g. SWIFT, Citibank NY, etc.)

39 Financial Stability and Interacting Networks of Financial Institutions and Market Infrastructures Seminar on Network Analysis and Financial Stability Issues Mexico City, Mexico, December 10 and 11, 2014