Interconnections between the French asset management sector and the rest of the French financial system

Transcription

1 Interconnections between the French asset management sector and the rest of the French financial system Kheira Benhami 1, Caroline Le Moign 2, Dilyara Salakhova 3, Alexandre Vinel 4 This study was prepared under the authority of the High Council for Financial Stability and does not necessarily reflect the position of the High Council or of the institutions whose authors are affiliated to. 1 Autorité des marchés financiers, Analysis, Strategy and Risks Division. 2 Autorité des marchés financiers, Analysis, Strategy and Risks Division. 3 Banque de France, Macroprudential Policy Division. 4 Working at the Directorate-General of the Treasury, Financial Sector Economic Analysis Division, during the realization of this work. Acknowledgments: this work was initiated within a working group gathering the Directorate-General of the Treasury, the Autorité des marchés financiers, the Banque de France and the Autorité de contrôle prudentiel et de résolution. The authors would like to thank, in particular, Eric Vansteenberghe, Edouard Chrétien and Anna Malessan for their active collaboration and the Directorate General Statistics of the Banque de France, in particular Emilie Candus, Gisèle Fourel and Maxime Ponsart, for their assistance in providing and working with the data.

2 Summary The risks related to the structure of the financial sector and particularly the interconnections between its various components (banks, insurance and asset management companies) raise many questions, notably about their contribution to the propagation and amplification of risks. These questions pertain in particular to the role of asset management. The interconnections between asset management sector and the rest of the French financial system result, on the assets side, from the holding of securities issued by financial entities (banks, insurers, or other financial intermediaries) and, on their liabilities side, from the holding of fund shares by these same types of entities. Based on regulatory reporting data, this study provides an analysis of the main characteristics of the interconnections between asset management sector and the French financial system over The results show that the observed network of interconnections between asset management sector and the rest of the financial system can be characterised as follows. First, it is sparse. Second, it exhibits a small world structure with a relatively short shortest average distance separating the network players from one another. Third, the network is organised around hubs (insurance, banks and certain money market funds) with funds being essentially connected to the banking sector on the assets side and to the insurance sector on the liabilities side. Money market funds stand out as being the most exposed to the banking sector where the latter represents 89% of their assets. Fund liabilities show higher concentration than assets with a particularly significant exposure to entities of the same group. These results can be later used to model shock propagation and to assess how a shock can be transmitted by investment funds.

3 I. Introduction The expansion of the financial sector has led to the development of complex chains of intermediation and interrelations between banks, insurers, and investment funds. The 2007 crisis, during which these channels contributed to the amplification of shocks and contagion, revealed the need to better understand these intersectoral links. Due to potential threat to financial stability posed by these interconnections, a correct assessment of the extent of these interconnections is crucial for macroprudential authorities. The regulators first phase of analysis consisted in identifying the largest bank nodes likely to spread contagion, notably with the work performed to identify global systemically important banks or G-SIBs (Basel Committee on Banking Supervision (2011)). The aim now is to acquire a detailed understanding of the entire intermediation network and its interactions in order to be able to map the vulnerabilities generated by these interdependencies. Many questions pertain in particular to the role of asset management sector in this network. The risks related to the characteristics of funds or their management (liquidity risks, risks associated with leverage, risks related to the assets held, operational risks, etc.) are already a core focus of regulatory action. In particular, instruments have been developed in a primarily microprudential approach (i.e., focusing on limiting risk or reinforcing the robustness of each player viewed in isolation). The risks related to the sector s structure and the interconnections, a macroprudential view, are difficult to assess as long as the links between players are not clearly identified. First, asset management interconnections may come from the funds assets. Among other things, the impact of these interconnections is greater when a fund or a family of funds hold a larger share of a market segment. Their behaviour may affect the market operation and other players in this market (investors, intermediaries, issuers). This is particularly the case in times of stress when a simultaneous and massive attempt to sell assets could affect asset prices and generate significant losses for the players. Interconnections arise also from the liabilities of these funds. They may affect counterparties who have invested in these funds but also counterparties exposed to these funds by other means (through the provision of credit lines or other financial services, counterparties on repo or derivative markets, etc.). In addition, unless this risk is properly managed, the counterparty risk can be amplified by the firstmover advantage, i.e., the exit of some investors at the beginning of a stressed market situation, leaving the remaining investors to bear the negative consequences. Lastly, interconnections may result from broader links (ownership relationships, sponsorship) that may lead another financial institution to provide support to a fund in difficulty because of a risk (for example, reputation risk) that it may face in case of the fund s liquidation. In this context, this study intends to shed light on interconnections involving the French asset management sector, using data received from the Banque de France, the ACPR (Autorité de contrôle prudentiel et de résolution), and the AMF (Autorité des marchés financiers) within the framework of regulatory reporting between 2008 and More specifically, we are interested in links between asset management sector and other institutions of the French financial system, i.e., banks and insurance companies domiciled in France. We have granular data allowing us to perform an analysis of the network at the level of individual entities (banks, insurance companies, and funds). 5 Due to the lack of data, discretionary management is not included in the analysis.

4 The assets of the French funds in our sample reach 1,450 billion euros at the end of Analyses show that the financial sector (banks, insurance companies, and funds) represent the main investors and investment destination of collective investment undertakings (CIU) domiciled in France. At this date, 51% of French CIU assets (712 billion euros) are invested in securities issued by entities in the financial sector as a whole, including 28% by entities domiciled in France (390 billion euros). The French entities held are mainly banks. The holdings are largely concentrated in certain types of funds, such as money market funds (350 billion euros in assets), with 88% of their assets invested in securities issued by financial institutions (47% of which are French institutions). In liabilities, 63% of CIU assets (915 billion euros) are held by financial entities, including 53% by French entities (770 billion euros), mainly insurers. Nevertheless, exposure to the financial sector has declined since The detailed analysis of interconnections resulting from the exposure of the French banking and insurance sectors to the asset management sector highlights the existence of a vast but relatively sparse network of 9,943 entities (funds, banks, and insurance companies) connected through 61,117 links (only links involving funds). The network is characterized by a small world structure, where network participants tend to form small groups among each other, while being accessible through a small number of links. This can be explained by the existence of hyper-connected hubs between them to which small entities in the network tend to connect. On the funds asset side, these entities are banks and money market funds. On the liabilities side, they are mainly insurers. Nevertheless, the analysis of concentrations of fund assets and liabilities shows that only 491 entities representing a total of 35 billion euros (2.4% of total assets) have significant levels of exposure to the French financial system in their assets. In liabilities, the number of funds concerned is 232 for a total of 38 billion euros (2.6% of total assets). Lastly, in order to have the most exhaustive analysis of interconnections between asset management and the rest of the French financial system, we look at the links between management companies and banks or insurance companies as being part of the same group. It turns out that a large part of the amounts invested by banks and insurance companies in French funds are invested with funds of the same group. These amounts represent 1% to 6% of the group s assets. The largest intra-group exposure is 60 billion euros in our network, less than 4% of the assets of the entity in question. The most exposed entity is an insurance company, with 6% of its assets representing 18 billion euros. After a detailed review of the academic literature and the description of the data, we examine the evolution of the asset management sector s exposure to the French financial institutions between 2008 and 2016 in the fourth section. We finish with a summary of the main characteristics of the interconnection network between asset management and the French financial system as of the end of 2016 in the last section. II. Literature review From a theoretical point of view, interconnections among financial institutions may have stabilising effects, resulting from the diversification of financial risks, but also amplifying effects, resulting from the creation of additional channels of shock propagation. Existing studies, focused on the banking sector, highlight the non-monotonous nature of interconnections on the system resilience: increasing the number of connections first increases and then decreases the probability of a liquidity crisis;

5 however, in the event of a major crisis, interconnections amplify the contagion mechanisms (Allen and Gale, (2000) or Gai, Haldane, and Kapadia, (2011)). The empirical economic literature on financial networks is rich in analyses aimed at better describing the nature and extent of interconnections, notably because of the introduction of this dimension in the methodologies used to identify systemic financial institutions (for banks and insurance companies, see BCBS 2013, IAIS 2013). Nevertheless, no consensus on the best way to quantify these interconnections has emerged. A first stream of analysis is that of a core-periphery structure and its impact on financial stability: Craig and von Peter (2014) analyse the German interbank network from 1999 to 2007 and identify a core of 45 banks among the 1,802 German banks, which are the largest banks, offering both retail and wholesale financing services. Their analysis allows a distinction between monetary core banks and others, with the former being larger contagion spreaders. Based on Dutch data from 1998 to 2008, van Lelyveld and In t Veld (2014) identify 15 central banks out of 100 active Dutch banks. They also observe that these banks are the least capitalised. Studying the US fed funds market from 1997 to 2006, Bech and Atalay (2010) observe a sparse small world network where bank s central position in the network predicts the interest rate that it offers on the interbank market. Puhr, Seliger, and Sigmund (2012) point out that Katz s centrality, i.e., the relative importance of a bank measured by the volume of its loans with other banks, is an indicator of systemic risk. By performing default simulations on Austrian bank data from 2008 to 2011, they observe that the contagion increases according to centrality, making it a good risk indicator. This literature highlights that the main nodes have their own characteristics and that their centrality can be an indicator of systemic risk. A second, more recent empirical stream of analysis aims to better understand the complexity of interbank relations by developing models with multiple layers of relations (or multiplexes ). Bargigli et al. (2013) examine the Italian banking market from 2008 to 2012 and show that the properties of the various layers (established according to the maturity of loans or provided collateral) have different topological properties (density, size, volumes) and that these characteristics are persistent. Extending these analyses for the network of major European banks at the end of 2011, Aldasoro and Alves (2017) conclude that the core-periphery analysis is not sufficient to understand the importance of a network agent due to the critical role of each agent within the multiplex. Therefore they propose new measures of systemic risk, which take into account institution s contribution examined at each layer of the multiplex. The authors thus identify around ten banks, going beyond the notion of centrality presented above, and point out in detail the components of their balance sheet that could be risk factors. However, most of the empirical studies characterising financial networks pertain to the banking sector, and only a few recent studies look at the nature of links between investment funds and the rest of the financial system. For example, Xiong and Nadal de Simone (2016) use portfolio data of investment funds and banks from the Central Bank of Luxembourg from 2008 to 2015 to study the impact of shock scenarios on variables used to assess systemic risk. They observe an evolution of interdependence between funds and banks over time, as well as an asymmetry in interconnections, with funds presenting a greater source of contagion for banks than the other way around. Gil-Bazo, Hoffmann, and Mayordomo (2017) observe, on the basis of portfolio data of Spanish funds from 2008 to 2012, how funds that have a capital relationship with a bank provide the latter with significant financing support, through their purchases of bonds on the primary market. This support, estimated at

6 a financing surplus of 514 million euros per bank (14 billion euros over the studied period), increased in times of financial stress and when funds are offered to individual investors (+70% bond purchases). Lastly, the ESRB (Abad et al. 2017) recently sought to use data collected by the European Banking Authority in its writing of guidelines on banks exposure to shadow banking. They thus have the exposure of 131 major European banks to all non-bank or insurance entities, for 560 billion euros (3,182 significant exposures, i.e., representing more than 0.25% of the capital reserve of the studied banks). 6 The interconnections studied here are limited to the asset or off-balance sheet holdings of European banks, which are included in the calculation of capital reserve requirements. Five countries account for 90% of exposures in value: the United Kingdom ( 284bn, 15% of GDP), Germany ( 106bn, 3.9% of GDP), France ( 78bn, 3.8%), Italy ( 27bn, 1.7%), and Luxembourg ( 10bn, but 23% of its GDP). The exposure of European banks is also concentrated in the types of links, since 65% of exposures concern three types of entities: securitisation vehicles (26%), non-money market investment funds (22%), and finance companies (18%). The interest in this database also lies in the possibility of studying cross-border interconnections. It shows that 60% of exposures are to entities outside the European Union, including 27% to US entities. Going beyond investment funds, this study thus presents the important role of other players in the network. Few studies of interconnections exist at this stage on French data: Hauton and Héam (2016) observe the effectiveness of several interconnection measures on the data of French banks and insurers in This study therefore complements the study presented here, which does not analyse the links between banks and insurers of the French network. They observe that the distribution of exposures is not uniform, with an important role played by conglomerates and very few exposures to insurers (which would be more exposed to their policyholders). They then look for new measures to analyse connectivity between these network agents (systemic importance, systemic fragility, credit risk), measures that are not easily transposable to asset management, concluding that these measures must be combined with contagion stress tests in a supervisory perspective. Heipertz et al. (2017) use aggregated national accounts data from 2000 to 2015 to assess a general equilibrium model, still in an exploratory stage. The model aims to describe the level of diversification of the asset and liability side of agents in the financial system, according to the demand for financial instruments. It also studies the effects of shock propagation resulting from changes in the price of these instruments. A more granular version, analysing these agents balance sheets line by line, could contribute to study propagation mechanisms in the French financial system. III. Data and methodology The scope of the analysis covers French collective investment undertakings (CIU) and their exposure to French financial institutions 8 over the period from 2008 to 2016, for which we have detailed data on assets and liabilities. 6 Money market funds (UCITS and AIF), investment funds, finance companies, broker-dealers, credit insurers, securitisation vehicles, vehicles similar to banks/insurance companies, and others. 7 Six financial conglomerates, four banks, and eleven insurers, representing more than 85% of the French financial sector. The institutions declare bilateral exposures totalling 227 billion euros, 90% of which are composed of debt instruments. 8 For their establishments domiciled in France. The available aggregate data also allow measuring the exposure to the entire financial system (i.e., including entities outside of France) over the same period.

7 On the asset side, Banque de France collects monthly CIU s investment portfolios at a security level. The DTOM and PROTIDE collections conducted with resident account-keepers provide detailed information on CIU securities held by banks. Lastly, the banking and insurance supervisor s TCEP reports provide the information on insurers holdings of CIU securities. 9 We therefore have detailed data on the assets of 10,241 funds representing nearly 1,450 billion in assets as well as data related to the holdings of French fund shares by 129 banks and 130 insurers as of The relationships between asset management and the banking and insurance sectors are represented by stocks and bonds issued by banks and insurance companies on funds asset side; and by fund shares held by banks and insurance companies on the funds liability side. Note that financial relationships other than those involving the securities holdings (credit, contingent credit lines, use of derivatives) are not captured in the analysis. In addition, for insurance entities, we make no distinction between securities held through euro or unit-linked accounts. However, it should be noted that for unit-linked contracts, the risks arising from holding assets are not borne by the insurers but by the end-investors, unlike securities held in euro accounts. At the sector level, the market share of unit-linked contracts is estimated at 11.6% of total assets in Our analysis consists of two parts. In the first one, we examine the evolution of the asset management sector s exposure to French financial institutions at an aggregate level. In the second part, we characterise the interconnections at the level of individual entities, i.e., funds, banks, and insurance companies, using concepts of network analysis. IV. Aggregate exposure of the asset management sector to the French financial system A. Continuous growth in assets of the French CIUs The assets of French funds increased significantly over the studied period, particularly after the end of the crisis, and grew 20% from 2011 to 2016 reaching 1,450 billion euros. For comparison, these assets are equivalent to 60% of the assets of French insurers and 20% of the assets of French banks. Graphique 1 - Assets of French CIUs However, this increase in assets, due to both valuation effects and net inflows, differ by asset class. Money market funds saw an overall decrease of nearly 6% of their assets between the end of 2011 and 2016, however, they started growing again over the last two years reaching nearly 350 billion euros. The other classes grew progressively over the same period, notably equity funds by 30% and bond funds by 33%. Source: data from Banque de France, AMF, ACPR 9 These detailed data are available only from 2010, the date of implementation of the collection.

8 As of 2016, the assets of French funds are allocated among funds as follows: 21% in equity funds, 18% in bond funds, 24% in money market funds, 23% in mixed funds, and 14% in other funds. 10 This growth in assets was also accompanied by a drop in the total number of funds from just over 11,000 funds at the end of 2011 to 10,241 at the end of 2016, which would partly reflect the streamlining of product ranges offered by portfolio management companies in order to increase clarity for investors. 11 B. An exposure to the financial system that tends to decrease Examining the global exposure data of French funds allows us to show their exposure to the entire financial system. 12 On the asset side, we distinguish between financial assets issued by entities domiciled in France, in the euro area, and outside the euro area. On the liabilities side, only financial entities in the euro area holding fund units can be identified in addition to French entities. Non-euro area financial entities are included in the generic category other. Graphique 2 - Exposure of fund assets to the rest of the financial system (in billions of and as %) Graphique 3 - Exposure of fund liabilities to the rest of the financial system (in billions of and as %) Source: data from Banque de France, AMF, ACPR. The differences between assets and liabilities can be explained by the lack of data on deposits and loans. As of the end of 2016, 51% of French CIU assets (712 billion euros) are invested in securities issued by entities in the financial sector as a whole, including 28% by entities domiciled in France (390 billion euros), 17% in the euro area excluding France, and 6% outside the euro area. On the liabilities side, 63% of CIU assets (915 billion euros) are held by financial entities of the euro zone, including 10 The other category mainly includes employees savings funds (60% of the category s assets) and private equity funds (retail private equity investment funds (FCPR), retail venture capital funds (FCPI), 39% of the category s assets). 11 Source: AMF, 2016 asset management key figures. 12 For the purposes of this study, the overall financial system groups together all financial institutions (or entities) (i.e., banks, insurance companies, funds, as well as any other financial institution) across all geographical areas. We will refer to the French financial system when these entities are domiciled in France.

9 53% by French entities (770 billion euros) and 10% by entities located in the euro area excluding France. On the asset side, the French entities concerned are mainly banks (17% of assets), but also units of other financial institutions (10% of assets). The liabilities are held by French insurers (33%), by other financial institutions (14%), and by French banks (6%). Also note that the increase in assets of French CIUs observed since 2012 was accompanied by a decrease in the exposure of French CIUs to the overall financial system. Although this decrease was continuous until the end of 2016 in CIU assets, the exposure of CIU liabilities to financial entities increased again between 2015 and This decrease is also not uniform: on the asset side, while the decrease in the share of securities issued by French financial entities was 7 points, there was an increase in the exposure to financial entities in the euro area and outside the euro area (+3 points since 2012). Lastly, on the liabilities side, the increase observed between 2015 and 2016 also resulted from entities in the euro area whose holding of units of French funds increased by 5 points over the same period. Nevertheless, French funds remain mainly exposed to French entities for funding. C. Mixed exposure to the financial system according to the type of fund In order to analyse exposure by fund type, we rely on the data available as of the end of December On the asset side, money market funds are unsurprisingly the most exposed to the overall financial system, which represents 88% of their assets, i.e., 270 billion euros. Of these investments, 47% are made in securities issued by French financial counterparties, 29% by entities in the euro area, and 24% by entities outside the euro area. Asset exposures to the entire financial system in other classes range from 20% for equity CIUs (80 billion euros) to 51% for mixed (balanced) CIUs (165 billion euros). It is reduced to 13% for equity CIUs and 29% for mixed funds, when only French financial institutions are considered. Graphique 4 - Exposure of fund assets to the rest of the financial system (in billions of ) Graphique 5 - Exposure of fund liabilities to the rest of the financial system (in billions of ) Source: data from Banque de France, AMF, ACPR, data for Q4 2016

10 On the liabilities side, the exposure appears less diverse: subscriptions of equity fund shares by financial counterparties amount to 73% of their liabilities (225 billion euros, including 190 billion euros by French entities). They reach almost the same proportion for bond CIUs at 180 billion euros (including 133 billion euros held by French entities). The exposure of money market CIUs is 230 billion euros (including 190 billion euros to French entities). Only the other category stands out with a 28% exposure to the financial system, including 26% to the French financial system in connection with the high proportion of employee savings funds in this category. V. Characterisation of the interconnection network between asset management sector and the rest of the French financial system A. The network is sparse and structured around hubs The interconnections between asset management and the rest of the French financial system result, on the assets side, from the holding of securities (equities and bonds) issued by financial entities (banks, insurers, or other financial intermediaries) and, on their liabilities side, from fund shares held by these same types of entities. To better understand the interconnections, the proposed analysis is based on concepts developed within the framework of network theory. A network consists of nodes and links. In our case, a node is a banking institution (B in Figure 6), an insurance institution (A), or a fund (F). We consider only French banks and insurance companies, and funds domiciled in France. In Q4 2016, the number of nodes in the network was 9,943, including 9,520 funds, 129 banks, and 130 insurance companies 13 (Figure 7). It declined since 2010, when there were 11,500 entities, due to the decrease in the number of funds domiciled in France observed over the period. These nodes are connected by links. It should be noted that all the links analysed in this study necessarily involve a fund. We thus do not study the existing direct links between banks and insurance companies. The existing links between two funds are included in the analysis. These links are thus materialised (i) on fund asset side, through holdings of stocks and/or bonds of credit institutions or insurers as well as holdings of fund shares and (ii) on fund liabilities side, through fund shares held by credit institutions, insurers or other funds. For greater clarity, the links will always be presented from the perspective of the fund. Graphique 6 - Graphic representation of the network Note: our network is composed of links between banks (B), insurance companies (I), and funds (F). These links are made to fund assets (link from F to I or B) or to fund liabilities (link I or B to F). We do not examine the links between I and B. 13 1,730 equity funds, 1,101 bond funds, 2,976 mixed funds, 14 hedge funds, 2,877 other funds, 234 money market funds, 129 banks, 130 insurers, and a residual balance of non-identifiable financial system entities (banks or insurers and other financial intermediaries).

11 As of Q4 2016, the network consisted of 61,117 links (Figure 7). This number appears to be limited compared with the 98 million potential links if all nodes were connected to a fund. It also decreased significantly since 2010, when there were 77,900 links. This decrease was particularly significant starting in The network s density, which is equal to the ratio of the number of existing network links and the number of possible links, appears to be very low with a level of 0.06%. This low density is in line with the results obtained in the empirical network analyses, which reveal a decreasing relationship between the number of links and the density of a network. For comparison, Anand et al. (2017) analyse different types of networks such as payment and interbank networks. Their results show that the density of payment networks reaches 0.6% for 5,733 nodes and the density of interbank networks reaches 1.1% for 535 nodes. Graphique 7 - Number of nodes and links of the network Graphique 8 - Distribution of the number of links of funds on their asset and liability side, by entity >1000 >100 & <1000 >10 & <100 <10 Assets Liabilities Source: data from Banque de France, AMF, ACPR Source: data from Banque de France, AMF, ACPR, data for Q The studied network is a directed network where the direction of the relationship between two entities matters: a link from entity B to entity F is different from a link from B to A. For example, a bank B holding units of fund F and fund F holding shares of bank B. It appears that the network links are essentially unidirectional, i.e., the links are made either to fund assets or to fund liabilities, and the proportion of bidirectional links is very low at 0.48%. The existing links are distributed very unevenly across entities. Figure 8 shows that the vast majority of network entities (mainly funds) have fewer than 10 links on either asset or liabilities side. Only 6 entities (insurance companies) hold shares of more than 1,000 funds and 3 entities (banks) are held by more than 1000 funds. This network structure assumes the presence of a few entities very connected with the rest of the system, and this structure is stable over time. The studied network has the characteristics of a vast directed network with low density. In addition, it presents structural characteristics typical of small world networks (small world properties) in network theory:

12 - The path between two network entities is limited, so that any participant in the network can be reached through a small number of links. The shortest path, which corresponds to the length of the shortest path in number of links between two network nodes, varies from 1 to 14 and averages These levels are relatively low given the number of entities that make up the network. For example, Boss et al. (2004) find that the average shortest path is equal to 2.6 links for 883 banks in the Austrian banking system. Bech and Atalay (2008) find an average shortest path of 2.4 links for a network of 479 fed funds. - On the other hand, the network s clustering, which measures the tendency of two entities connected to a third to be connected to each other, is 12%, a value much higher than what a random network with similar characteristics (in number of nodes and links) would have, which is around 0.1%. This confirms a small world property, where network participants tend to form small groups among each other, while being accessible through a small number of links. These two parameters were relatively stable since Lastly, assortativity, which measures the propensity of two entities with a large number of links to be linked together, is a negative value of in the case of our network. The network is therefore disassortative: entities with few connections tend to be connected to entities that have a more developed network. In historical terms, this dissertativity increased slightly since This result highlights the fact that the network is organised around hubs to which small entities of the network (i.e., having a small number of links) are connected. Graphique 9 - Average shortest path, assortativity, and clustering Source: data from Banque de France, AMF, ACPR The rest of the presented work will focus on a detailed analysis of the characteristics of the network and its constituent entities. For the sake of clarity, the analyses are conducted on the latest available data, in Q B. A management sector relatively concentrated around a few management companies The French asset management sector has a relatively large number of asset management companies (630 at the end of 2016). However, it appears that the top ten asset management companies account for nearly 60% of total net assets, of which the three largest asset managers represent 20%, 8%, and 8% of total net assets respectively and whose number of funds in our base reaches 2,000, 1,200, and 1,100

13 respectively. We are interested in the characteristics of the sub-networks formed by funds of the same management company with their counterparties. Sub-networks of management companies consist of all funds belonging to the same management company as well as the entities to which they are linked and the links that connect them in the network. Figure 10 details the characteristics of the links of the sub-network formed by the 10 largest management companies with the rest of the French financial system. It turns out that the number of links per management company changes proportionally with the total number of funds that make it up: the largest entity has a total number of nearly 7,000 links for 2,000 funds compared with fewer than 900 links for the smallest of the ten, with 270 funds. The same proportionality is found in the analysis of the total amounts committed in the network. The largest entity thus accounts for 164 billion euros in the network (total assets and liabilities). However, this does not signal a possible concentration of those amounts, within a management company, on certain funds in particular. For the top ten asset management companies, the share of assets exposed to the French financial system ranges from 27% to 43%. Graphique 10 - Top 10 management companies Source: data from Banque de France, AMF, ACPR, data for Q Note: each bubble represents a management company. The size of the bubbles is proportional to the assets managed by the management company. C. Banking entities highly present in fund assets, insurers and money market funds in their liabilities As the aggregate analysis showed, funds are exposed to French banks on the asset side and to French insurers on the liabilities side. An average bank is connected to 152 funds, or in other words, securities issued by an average bank are held by 152 funds, with an average total amount of holdings reaching 1,931 million euros. On the liabilities side, an average insurer holds shares of 151 funds for an average total amount of 3,409 million euros.

14 Graphique 11 - Average number of links with funds per entity Graphique 12 - Average total value of links per entity (in millions of ) Source: data from Banque de France, AMF, ACPR, data for Q Note: French banks are exposed on average to 152 funds by their presence in the asset side for an average total amount of 1.9 billion euros per bank. French insurers are exposed on average to 151 funds (by their presence in their liabilities side) for an average total amount of 3.4 billion euros per insurer. Money market funds (350 billion euros in assets for 234 entities) have the highest intensity of interconnections with the rest of the network, with an average of 8 links (or 565 million) on the asset side and 38 links ( 817 million) on the liabilities side. Other types of funds (bonds, mixed, money market, and others) have a significantly smaller number of links, particularly on the liabilities side. The difference is even greater when the average amounts of links are considered. Graphique 13 - Average number of links in assets and in liabilities, by type of fund Graphique 14 - Average total value of links in assets and in liabilities, by type of fund, (in millions of ) Source: data from Banque de France, AMF, ACPR, data for Q D. Funds are more exposed to hubs on their liabilities side The ability to reconcile a network with low density and a short average path between its nodes is explained by the presence of a few hubs. The latter are essentially banks, insurance companies, and money market funds with multiple links. This characteristic of the network is confirmed in figures 15 and 16.

15 Graphique 15 - Top 10 entities links on the funds liabilities side Graphique 16 - Top 10 entities linkson the funds asset side Source: data from Banque de France, AMF, ACPR, data for Q Note: The red colour corresponds to insurers, the green colour to banks, and the orange colour to money market funds. The largest network entity on the liabilities side invested 67 billion euros in the French asset management sector through 680 links. French CIUs hold 23 billion euros in assets of the top banking entity through 1,878 links. The top ten entities on the funds liabilities side invest 310 billion euros in the shares of French funds, representing 22% of funds total net assets. These entities are almost exclusively insurance companies. However, note the presence of a money market fund that invests 14 billion euros in fund shares. In addition, the top two entities account for almost 120 billion euros of fund liabilities through a relatively small number of links (680 and 926 links respectively). More generally, for the top ten entities, the average size per entity in fund shares is 37 billion euros. On the asset side, the ten largest holdings of French CIUs account for a total of nearly 120 billion euros (9% of total assets). They consist of banks (60% of total amounts) but also of money market funds (40% of amounts). The amounts per entity on the asset side are significantly lower than those of the top entities present on the liabilities side: the amounts held for the top two entities total 42 billion euros, and the average size per entity of the top ten entities on the asset side is 18 billion euros, i.e., half of that on the liabilities side. The results therefore suggest that fund liabilities are more concentrated than fund assets around larger hubs. Charts 1 and 2 in the appendix provide a graphic view of the links between these entities. In order to be able to assess the intensity of fund s exposure to the French financial system, we look at the degree of concentration of fund assets and liabilities. This analysis is based on the Herfindahl index, which measures the degree of concentration of the assets (or liabilities) of each fund. A high Herfindahl index of assets implies that a fund has mainly invested in securities issued by a small number of entities. Symmetrically, a high Herfindhal index of liabilities corresponds to a low diversification of fund investors. We are especially interested in funds that are exposed to the French financial sector. The Herfindahl index is therefore cross-tabulated with the share of assets invested in assets issued by French financial institutions (or the share of fund liabilities held by French financial institutions) (Figures 17 to 20).

16 Graphique 17 - Analysis of the concentration of fund assets (in number of funds) Graphique 18 - Analysis of the concentration of fund assets (in amounts, in millions of ) Source: data from Banque de France, AMF, ACPR, data for Q Graphique 19 - Analysis of the concentration of fund liabilities (in number of funds) Graphique 20 - Analysis of the concentration of fund liabilities (in amounts, in millions of ) Source: data from Banque de France, AMF, ACPR, data for Q4 2016

17 Making the strong assumption that a fund is highly exposed to the French financial sector when more than 50% of its assets are invested in securities issued by entities of the French financial sector (or more than 50% of its liabilities are held by French financial entities), it appears that only 491 entities representing a total of 35 billion euros (2.4% of total assets) have significant levels of exposure to the French financial system on the asset side. On the liabilities side, the number of funds concerned is 232 for a total of 38 billion euros (2.6% of total assets). E. Are funds more exposed to banking and insurance entities of the same group? In order to have the most exhaustive analysis of interconnections between asset management and the rest of the French financial system, we look at the links that exist beyond the direct holding of assets. In particular, we look at the links between management companies and banks or insurance companies resulting from being part of the same group. During the financial crisis, it appeared that several financial institutions (in particular, banks) supported troubled entities, particularly in asset management, and that in a number of cases, these interventions ( step-in ) even took place when no contractual commitment required them to do so. 14 These risks are now regulated at the international level, through the introduction of regular analyses of the risk of implicit or explicit support by banks and regulators, which may lead to additional capital requirements, if applicable 15. Figures 21 and 22 thus provide an analysis of the exposures of the top ten asset managers to the top ten banking and insurance groups in their assets and liabilities. In other words, these are, respectively, the amounts invested by the top ten management companies in securities issued by the ten largest banks and insurers and the amounts that were invested by the ten largest banks and insurance companies in the funds of the ten largest management companies. It should be noted that, for a given management company, only exposures through its French funds are included in the analysis. A fund of a French management company domiciled in Luxembourg will thus not be taken into account in this analysis. 16 Exposures between entities in the same group are shown in red both for assets and liabilities sides. 14 Moody s estimates that around 60 funds (out of 1,091 funds) analysed received support worldwide. See Moody s Investors service (April 2016), Basel Committee Seeks to Capture Step- In Risk Involved In Asset Management Activities, More Capital Charges Ahead,. 15 See Basel Committee on Banking Supervision (October 2017), Identification and management of step-in risk,. 16 Assets of European collective investment undertakings (CIUs) managed by French management companies are estimated at 316 billion euros in 2016 (source: AMF, 2016 asset management key figures).

18 Graphique 21 - Exposure of assets of the top French management companies to the top French financial institutions (in millions of euros) Graphique 22 - Exposure of liabilities of the top French management companies to the top French financial institutions (in millions of euros) Source: data from Banque de France, AMF, ACPR, data for Q Note: The red colour corresponds to exposure to a banking or insurance entity of the same group. The first finding is that on the asset side, funds exposures to entities of the same group are similar to exposures to other entities and do not present any imbalance. It is not surprising since this diversification stems from the microprudential management rules imposed on funds. For example, an undertaking for collective investment in transferable securities (UCITS) may not invest more than 5% of its assets in securities or money market instruments issued by the same entity. On the other hand, on the liabilities side, large amounts invested by banks and insurance companies in French funds are invested in funds of the same group. To assess the significance of these exposures, we report them in terms of the groups consolidated assets. The amounts invested in funds by the largest banks and insurance companies thus represent 1% to 6% of their total assets. The largest intragroup exposure amounts to 60 billion euros in our network, less than 4% of the assets of the entity concerned. The most exposed entity (an insurance company) has an exposure of 6% of its assets representing 18 billion euros. VI. Conclusion The results show that the network currently formed between asset management and the rest of the financial system is sparse. The mapping carried out also reveals a small world structure, organised around hubs (insurance banks and certain money market funds) where the average distance separating the network s players from one another is relatively short. Funds are essentially connected to the banking sector through their assets and to the insurance sector through their liabilities. We observe that money market funds are the most exposed with 89% of their assets consisting of bank assets. Fund liabilities also remain more concentrated than fund assets and show a significant exposure to entities of the same group. The detailed characterisation of the interconnections between the French asset management sector and the other participants of the French financial system presented in this study is an important contribution. While the theoretical literature on contagion contributes to a better understanding of the risks related to interconnections, the empirical literature, particularly on European or French data, can be further enriched. It would therefore be interesting to have similar characterisation for other countries.

19 The results can be incorporated into shock propagation models to assess how, in this case, a shock can be amplified or otherwise mitigated through exposures to investment funds. Nevertheless, this part remains nascent, and it is necessary in stress test analysis. to consider not only interconnections but also regulatory constraints imposed on the asset management sector (in terms of asset diversification, asset liquidity constraints, possibility of redemptions, etc.) and the tools available to managers to manage liquidity in the event of stress (redemption capping mechanisms or gates, subscription and/or redemption notice periods, redemptions in kind, and partial or total closure of subscriptions in particular). More generally, the advantages or disadvantages of the current situation may be considered in view of a major counterfactual: the absence of investment funds, which would result in a direct ownership of assets by the players in question. Finally, other areas of work could also address how a network structure interacts with other potential sources of contagion (for example, level of leverage, capital buffer, credit trends, short-term financing, etc.). What is its role? In this context, is it a factor of propagation or mitigation of systemic shocks? What are the characteristics of the network that could possibly be taken into account in the prevention of contagion mechanisms? An empirical analysis could illustrate a theoretical literature such as that of Eisenberg and Noe (2001), who model loss transmission mechanisms, or Glasserman and Young (2016), who examine the characteristics of the network and their contribution to systemic risks (vulnerability of agents, assets held, use of leverage, common exposures, etc.).

20 Bibliography Abad, J., M D Errico, N Killeen, V Luz, T Peltonen, R Portes and T Urbano (2017), Mapping the interconnectedness between EU banks and shadow banking entities, European Systemic Risk Board Working Paper No. 40. Aldasoro I. and Alves, I. (2017), Multiplex interbank networks and systemic importance - An application to European data, BIS Working Papers No Allen, F. and D. Gale (2000), Financial contagion, Journal of political economy, 108(1), Basel Committee on Banking Supervision (2011), Basel III: A global regulatory framework for more resilient banks and banking systems, Basel Committee on Banking Supervision. Basel Committee on Banking Supervision (2013), Global systemically important banks: updated assessment methodology and the higher loss absorbency requirement, Basel Committee on Banking Supervision Consultative Document. Bargigli, L., di Iaso, G., Infante, L., Lillo, F., Pierobon, F. (2013), The multiplex structure of interbank networks, arxiv: v1. Bech M., and E. Atalay, (2010), The Topology of the Federal Funds Market, Physica A: Statistical Mechanics and Its Applications 389 (22): M. Boss, H. Elsinger, M. Summer et S. Thurner (2004), Network topology of the interbank market, Quantitative finance, vol 4. Issue 6, Craig, B. and G. von Peter (2014), Interbank Tiering and Money Center Banks, Journal of Financial Intermediation 23 (3): Elsinger, H. (2009), Financial Networks, Cross Holdings, and Limited Liability, Oesterreichische Nationalbank Working Paper 156. Gai, P., A. Haldane and S. Kapadia (2011), Complexity, Concentration and Contagion, Journal of Monetary Economics 58 (5): Gil-Bazo, J., Hoffmann, P. and Mayordomo, S. (2017), Mutual Funding, SSRN. Glasserman, P., and Young, H.P. (1016), Contagion in Financial Networks Journal of Economic Literature vol. 54 (3): Hauton, G. and Héam J.C. (2016): How to measure interconnectedness between Banks, Insurers and Financial Conglomerates?, Statistics & Risk Modeling, 33, 3 4, Heipertz J. and Ouazad A., Rancière R. and Valla N. (2017), Balance-Sheet Diversification in General Equilibrium: Identification and Network Effects, National Bureau of Economic Research Working Paper No IAIS (2013), Global systemically important insurers: Initial assessment methodology, International Association of Insurance Supervisors. In t Veld, D. and van Lelyveld I. (2014), Finding the Core: Network Structure in Interbank Markets, Journal of Banking and Finance 49: Puhr, C., Seliger, R. and Sigmund M. (2012), Contagiousness and Vulnerability in the Austrian Interbank Market, Oesterreichische Nationalbank Financial Stability Report 24.

21 Xisong Jin X. and F. Nadal De Simone (2016), Tracking Changes in the Intensity of Financial Sector s Systemic Risk, Central Bank of Luxembourg working papers No. 102.

22 Appendix 1 Representation of the network from the held entity s point of view (liabilities side) The size of the network nodes is proportional to the number of links weighted by their amount. The size of the links is proportional to the amounts involved. The represented link goes from the holding entity to the issuing entity. The link colour corresponds to the type of holding entity. This representation allows us to see that the most held entities are banks (red) and money market funds (aqua). They are also held by money market funds (aqua links) and insurance companies (dark blue links) respectively. Figure 1 Source: data from Banque de France, AMF, ACPR, data for Q4 2016

23 Appendix 2 Representation of the network from the holding entity s point of view (asset side) The size of the network nodes is proportional to the number of links weighted by their amount. The size of the links is proportional to the amounts involved. The represented link goes from the holding entity to the issuing entity. The link colour corresponds to the type of holding entity. This representation allows us to see that the entities with the largest exposures on the asset side are essentially insurance companies (dark blue) and, to a lesser extent, money market funds (aqua). They mainly hold money market funds (aqua links) and banks (red links) respectively. Figure 2 Source: data from Banque de France, AMF, ACPR, data for Q4 2016