Risk Networks Sanjiv R. Das Santa Clara University @IRMC Warsaw - PowerPoint PPT Presentation

Risk Networks Sanjiv R. Das Santa Clara University @IRMC Warsaw June 2014 Sanjiv R. Das Risk and Return Networks IRMC 2014 1 / 47

Outline 1 A review of risk metrics on networks. 2 A big data application to interbank loan networks for banking systemic risk in the U.S., using text mining and network analysis. 3 A new approach to systemic risk on networks. 4 Risk and return on venture capitalist networks. Relevant papers: http://algo.scu.edu/ ∼ sanjivdas/vccomm.pdf http://algo.scu.edu/ ∼ sanjivdas/midaswww2011 FINAL.pdf Sanjiv R. Das Risk and Return Networks IRMC 2014 2 / 47

Part 1: Network Metrics Concepts and calculations Graph Theory: Network Types Node/Vertex (V) Edge (E) Degree (d) = 6 Network/Graph = G(V,E) f ( d ) ∼ N ( µ, σ 2 ) f ( d ) = d − α , 2 < α < 3 ’ Sanjiv R. Das Risk and Return Networks IRMC 2014 3 / 47

Part 1: Network Metrics Concepts and calculations Random vs Scale-Free Graphs Barabasi, Sciam, May 2003 Sanjiv R. Das Risk and Return Networks IRMC 2014 4 / 47

Part 1: Network Metrics Concepts and calculations Centrality (Bonacich 1987) Also known as PageRank by Google. Adjacency matrix: A ij ∈ R N × N Influence: x i = � N j =1 A ij x j λ x = A · x · Centrality is the eigenvector x corresponding to the largest eigenvalue. Centrality scores = {0.71, Centrality scores = {0.58, Centrality scores = {0.71, 0.50, 0.50} 0.58, 0.58} 0.63, 0.32} Sanjiv R. Das Risk and Return Networks IRMC 2014 5 / 47

Part 1: Network Metrics Concepts and calculations Diameter Longest shortest distance from a node to any other node, across all nodes. The diameter of this graph is 2. Sanjiv R. Das Risk and Return Networks IRMC 2014 6 / 47

Part 1: Network Metrics Concepts and calculations Fragility Definition: how quickly will the failure of any one node trigger failures across the network? Is network malaise likely to spread or be locally contained? Metric: R = E ( d 2 ) E ( d ) , where d is node degree. Fragile if R > 2. Fragility of the sample network = 20 Sanjiv R. Das Risk and Return Networks IRMC 2014 7 / 47

Part 1: Network Metrics Concepts and calculations Communities Definition: clusters of nodes that interact much more within community than across community. Hard computational problem. Fast-greedy algorithm (Girvan & Newman 2003) Walk-trap algorithm (Pons & Latapy 2005) Sanjiv R. Das Risk and Return Networks IRMC 2014 8 / 47

Part 1: Network Metrics Concepts and calculations Modularity Quasi-distance metric between community based adjacency matrix partition and one with no communities. Metric: Q = 1 � A ij − d i × d j � � � · δ i , j ( C k ) 2 m 2 m K i , j A ij m = � 2 . So, 2 m is the sum of all edges. i , j δ i , j ( C k ) = 1 if i , j are in the same community, else zero. Sanjiv R. Das Risk and Return Networks IRMC 2014 9 / 47

Part 2: Systemic Risk from Co-Lending Networks Defining systemic risk analysis Systemic Analysis 1 Definition: the measurement and analysis of relationships across entities with a view to understanding the impact of these relationships on the system as a whole. 2 Challenge: requires most or all of the data in the system; therefore, high-quality information extraction and integration is critical. Sanjiv R. Das Risk and Return Networks IRMC 2014 10 / 47

Part 2: Systemic Risk from Co-Lending Networks Defining systemic risk analysis Systemic Risk 1 Current approaches: use stock return correlations (indirect). [Acharya, et al 2010; Adrian and Brunnermeier 2009; Billio, Getmansky, Lo 2010; Kritzman, Li, Page, Rigobon 2010] 2 Midas: uses semi-structured archival data from SEC and FDIC to construct a co-lending network; network analysis is then used to determine which banks pose the greatest risk to the system. Sanjiv R. Das Risk and Return Networks IRMC 2014 11 / 47

Part 2: Systemic Risk from Co-Lending Networks The Midas Project Midas Project: Overview Joint work with IBM Almaden 1 Focus on financial companies that are the domain for systemic risk (SIFIs). Extract information from unstructured text (filings). Information can be analyzed at the institutional level or aggregated system-wide. Applications: Systemic risk metrics; governance. Technology: information extraction (IE), entity resolution, mapping and fusion, scalable Hadoop architecture. 1 “Extracting, Linking and Integrating Data from Public Sources: A Financial Case Study,” (2011), (with Douglas Burdick, Mauricio A. Hernandez, Howard Ho, Georgia Koutrika, Rajasekar Krishnamurthy, Lucian Popa, Ioana Stanoi, Shivakumar Vaithyanathan), IEEE Data Engineering Bulletin , 34(3), 60-67. [Proceedings WWW2010, April 26-30, 2010, Raleigh, North Carolina.] Sanjiv R. Das Risk and Return Networks IRMC 2014 12 / 47

Part 2: Systemic Risk from Co-Lending Networks The Midas Project Entity View Midas ¡provides ¡an ¡en-ty ¡view ¡around ¡new ¡sources ¡of ¡data ¡ • Extraction and cleansing of financial entities, their resolution and linkage across multiple sources Web Data • Uncovering non-obvious relationships between financial entities News ¡ Blogs ¡ • Computation of key financial metrics using data extracted from multiple sources of public data Reviews ¡ • Information analyzed at the institutional level or aggregated system-wide. FDIC ¡Call ¡Data ¡ Public Data Midas Records ¡ Financial SEC ¡Filings ¡ Insights OTS ¡Thri6 ¡ Financial ¡Records ¡ • Regulators • Credit committees • Investment analysts Private Data • Portfolio managers Hoovers ¡ D&B ¡ FINRA ¡ • Equity managers Private ¡Wall ¡Street ¡Journal ¡ Sanjiv R. Das Risk and Return Networks IRMC 2014 13 / 47

Part 2: Systemic Risk from Co-Lending Networks The Midas Project Input & Output Midas ¡Financial ¡Insights ¡ Insider Transaction Proxy Statement Annual Report Loan Agreement Raw ¡Unstructured ¡Data ¡ Raw ¡Unstructured ¡Data ¡ Extract Integrate Data ¡for ¡Analysis ¡ Exposure by subsidiary Related Companies Loan Exposure … ¡ … ¡ Sanjiv R. Das Risk and Return Networks IRMC 2014 14 / 47

Part 2: Systemic Risk from Co-Lending Networks The Midas Project Process Example ¡of ¡Midas ¡Financial ¡Insights ¡ Over ¡1 ¡Million ¡documents ¡ Filing ¡ Bmeline ¡ ¡ ¡ ¡2005 ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡ ¡2010 ¡ ¡ ¡ ¡ ¡ Over ¡32000 ¡key ¡officials ¡ in ¡financial ¡companies ¡ Filings ¡of ¡ Person Financial ¡ ¡ Companies ¡ Extract Integrate ¡ ( Forms ¡10-­‑K,8-­‑k, ¡10-­‑Q, ¡DEF ¡ Company Over ¡2200 ¡financial ¡companies ¡ 14A, ¡3/4/5, ¡13F, ¡SC ¡13D ¡SC ¡ 13 ¡G ¡ FDIC ¡Call ¡Reports) ¡ ¡ Call ¡Data ¡ Records ¡ SEC ¡Filings ¡ Sanjiv R. Das Risk and Return Networks IRMC 2014 15 / 47

Part 2: Systemic Risk from Co-Lending Networks Data Handling Data Midas ¡provides ¡Analy0cal ¡Insights ¡into ¡company ¡rela0onships ¡by ¡exposing ¡informa0on ¡concepts ¡and ¡ rela0onships ¡within ¡extracted ¡concepts ¡ Current ¡Events ¡ • merger ¡and ¡acquisi8on ¡ • bankruptcy ¡ • change ¡of ¡officers ¡and ¡directors ¡ Subsidiaries ¡ • material ¡defini8ve ¡agreements ¡ • list ¡subsidiaries ¡of ¡a ¡ company ¡ Forms 8-K Officers ¡& ¡Directors ¡ • men8on ¡ Forms 3/4/5, SC 13D, SC 13G, 10-K, FDIC Call Report • bio ¡range, ¡age, ¡current ¡ Event posi8on, ¡past ¡posi8on ¡ • signed ¡by ¡ subsidiaries, insider, 5%, 10% owner, banking • commiNee ¡membership ¡ Forms 3/4/5, SC 13D, SC 13G subsidiaries employment, director, officer Shareholders ¡ insider, 5% owner, 10% owner • related ¡ins8tu8onal ¡managers ¡ • Holdings ¡in ¡different ¡securi8es ¡ Forms 10-K, DEF Company Person 14A, 8-K, 3/4/5, 13F, SC 13D, SC 13G, FDIC Call Report borrower, lender holdings, Forms 10-K, DEF 14A, 8-K, 3/4/5 transactions Forms 10-K, 10-Q, 8-K Loan Security 5% ¡beneficial ¡ownership ¡ Reference SEC table • owner ¡ Forms 13F, Forms 3/4/5 • issuer ¡ Loan ¡Agreements ¡ • % ¡owned ¡ • loan ¡summary ¡details ¡ • date ¡ • counterpar8es ¡(borrower, ¡ Insider ¡filings ¡ lender, ¡other ¡agents) ¡ • transac8ons ¡ • commitments ¡ • holdings ¡ • Insider ¡rela8onship ¡ 7 ¡ Sanjiv R. Das Risk and Return Networks IRMC 2014 16 / 47

Part 2: Systemic Risk from Co-Lending Networks Data Handling Loan Extraction Example ¡Analysis ¡: ¡Extrac3on ¡of ¡Loan ¡Informa3on ¡Data ¡ Extract and cleanse information from headers, tables main content and signatures Id Company Role Commitment 1 Charles Schwab Corporation Borrower 1 Citibank, N.A. Administrative Agent Id Agreement Name Date Total Amount 1 Citibank, N.A. Lender $90,000,000 1 Credit Agreement June 12, 2009 $800,000,000 1 JPMorgan Chase Bank, N.A. Lender $90,000,000 … 1 Bank of America, N.A. Lender $80,000,000 … Loan Information Loan Company Information Notes: ¡ ¡Loan ¡Document ¡filed ¡by ¡Charles ¡Schwab ¡Corpora3on ¡On ¡Aug ¡6, ¡2009 ¡ ¡ Sanjiv R. Das Risk and Return Networks IRMC 2014 17 / 47