Statistically validated network of portfolio overlaps and systemic risk

Télécharger le fichier pdf d’un mémoire de fin d’études

Factset Ownership – General description

Factset is a financial data provider used by investment professionals (e.g. porftolio managers, analysts, investment bankers, fund managers) for their fund screening and selection process. FactSet provides multiple interconnectable datasets (e.g. FactSet Fundamentals, FactSet Estimates, Factset Ownership). We focused on FactSet Ownership in this thesis and will describe its scope and content in this chapter.
The FactSet Ownership database collects global equity ownership data for about 13 000 institutions, 33 000 fund portfolios, and 280 000 non-institutional stake-holders with history going back until 1999. FactSet aggregates publicly available sources (EDGAR forms 13F, N-O, N-CSR and occasionally 485BPOS) for its ownership data. Therefore every data collected is marked with a source name and a report date. Depending upon the nature of the source, the report date may be updated monthly, quarterly, semiannually or annually.
In addition, FactSet actively enriches its data by approaching mutual funds (mainly in Europe) individually to invite them to provide their data.

FactSet identifiers

I provide a short description of three identifiers (entity based, security based and fund based) generated and maintained by FactSet. They are all defined as unique and never changing with respect to their universe and make it possible to merge datas across the FactSet Ownership tables and also across other FactSet maintained databases.

FactSet Entity Identifier

It is the most general identifier defined by FactSet: it is defined as any organization or individual that is included in any of FactSet content sets. Since FactSet provides its dataset in a relational database format, it allows to seamlessly integrate all the tables together, not only in the Ownership database but also across all the datasets. Financial institutions are defined by their entity identifier.

FactSet Permanent Security Identifier

The security identifier is available under the name fs_perm_sec_id. The first 6 characters of the fs_perm_sec_id are a random alphanumeric combination, followed by “-S-” and lastly followed with a 2 character code representing the country in which the security trades. Therefore multiple security identifiers can correspond to a single entity identifier if it is traded in multiple exchanges, as shown in Table 2.1.

FactSet Fund Identifier

As opposed to the financial institutions which are assigned a unique entity identifier, funds have their own identifier: the FactSet Fund Identifier. The distinction between funds and institutions happen because multiple funds can be related to the same institution (e.g. BlackRock is an institution therefore is assigned an institution identifier, however its multiple funds are assigned a unique fund identifier each).

Notations

Let us define some necessary quantities to be more precise, the notations defined in this section will be sensibly valid through the whole thesis, unless said otherwise.
At quarter q, fund (resp. institution) i has capital W( f )
i (q) (resp. W(i)
i (q)) which is invested into n( f )
i (q) (resp. n(i)
i (q)) securities among M(q) existing ones. As a result, each security a, whose capitalization is denoted by Ca(q), is found in m( f )
a (q) funds’ (resp. m(i)
a (q) institutions’) portfolios. W( f )
ia (q) (resp. W(i)
ia (q)) is the position in dollars of fund (resp. institution) i on security a at the end of quarter q.
In all the forthcoming chapters of this thesis we will consider only funds or institutions separately (chapter 3 will focus on institutions whereas chapter 4 and 5 on funds), therefore, for the sake of clarity, we will drop the exponents f and i when there is no ambiguity. Time referring to quarterly updated data will be denoted by q and by t otherwise. However, the explicit time dependence will usually be dropped when not necessary.

FactSet – Securities

A security is a good or an asset that holds a monetary value. Its legal definition varies by jurisdiction: In the United States, a security is a tradable asset of any kind.
Securities are usually categorized into:
equity securities (common stocks)
debt securities (banknotes, bonds, and debentures)
derivatives (forwards, futures, options and swaps)
In the scope of this thesis, a security will usually refer to an equity security, that is, an ownership interest held by a shareholder in an entity (company, partnership or trust). Securities included in the database are actively traded and are one of the issue type defined in Table 2.2. As we explained in the introduction, FactSet is constantly increasing its range of coverage. As a result Figure 2.1 shows a sensible increase in the total number of securities and total market capitalization over the studied period of time. One should keep that property of the dataset in mind when studying… as it can have great consequences on, for example, the computation of average degree of a bipartite network.
Since securities are selected by institutional investors, they only play a passive role in the Ownership database. Therefore they only have two dedicated tables which are the “own_basic” defined earlier and the “own_prices” (see Table 2.2) which is a record of the price history and share outstanding of the securities. This table is useful to convert the number of shares into dollar, and therefore to compute the portfolio value of an investor, as well as the market capitalization of a security.
The vast majority, more than ninety percent, of the securities are located in North America, Europe and Asia.

Distributions

We plot the distribution of company size in term of market capitalization (fig. 2.6), number of institutional investors (fig. 2.9), number of fund investors (fig. 2.15) and stock prices (fig. 2.5). These distributions will be studied in greater details in section 2.9).

FactSet – Institutions

The primary source for institutional ownership of U.S.-traded equity is the 13F filing.
This filing is mandated by the SEC for any investment management institution managing $100 million or more in U.S.-traded securities. Approximately 3,200 institutions file 13Fs on a quarterly basis (filed within 45 days of the calendar quarter-end, and reporting holdings as of that quarter-end). These filings are available electronically on the SEC’s EDGAR system. Form 13F is limited to the EDGAR system. As the 13F requires to report only positions greater than 10000 shares or $200000 on listed securities, some positions are already filtered out. Also it represents only the aggregated positions of the institutions: the portfolios of sub-funds, that we will describe in the next section, are merged into a single report.
Using the 13F filing is convenient because all the financial institutions are required to report quarterly and at the same dates. Therefore the discrete time series are observed in a synchronous manner and, as we have seen, small positions are not reported so already filtered out.
Table 2.3 shows the structure of the table that contains all the history of the positions held by the institutions. Here, the factset_entity_id represents an institution. For example, the institution 002H53-E held 8100 shares of the security H0PJ8L-S-US as of 2011-12-31. From that table it is possible to reconstruct the portfolio of any institution in the database at any time.

Table des matières

1 Introduction
2 Datasets
2.1 Factset Ownership – General description
2.2 FactSet identifiers
2.2.1 FactSet Entity Identifier
2.2.2 FactSet Permanent Security Identifier
2.2.3 FactSet Fund Identifier
2.3 Notations
2.4 FactSet – Securities
2.4.1 Distributions
2.5 FactSet – Institutions
2.5.1 Distributions
2.6 FactSet – Funds
2.6.1 Non-synchronous time-series
2.6.2 Cleaning and filtering
2.6.3 Distributions
2.7 Thomson-Reuters Tick History
2.8 Thomson-Reuters Tick History and FactSet matching
2.9 Heavy tail distributions
2.9.1 Method
2.9.2 Results
3 Statistically validated network of portfolio overlaps and systemic risk
3.1 Results and Discussion
3.1.1 Temporal evolution of the validated network of institutions
3.1.2 Validated overlaps vs portfolio size and security capitalization
3.1.3 Distressed institutions in the validated networks
3.1.4 Buy and sell networks: the case of Hedge Funds
3.1.5 Temporal evolution of the validated network of securities
3.2 Discussion
3.3 Methods
3.3.1 Dataset
3.3.2 Significance level under multiple tests
3.3.3 Resolution problems for the hypergeometric distribution approach
3.3.4 p-values from the Bipartite Configuration Model
4 Collective rationality and functionalWisdom of the Crowd in far-from-rational institutional investors
4.1 Introduction
4.2 Wisdom of the Crowd
4.3 Asset selection model
5 Large large-trader activity weakens long memory of limit order markets
5.1 Introduction
5.2 The data
5.3 Methods
5.3.1 Microstructure: memory length of market order sign auto-correlation
5.3.2 Macro-dynamics: directional fund activity ratio
5.3.3 Macro-dynamics: absolute fund activity ratio
5.4 Results
5.4.1 From large fund behaviour to microstructure dynamics
5.4.2 Large fund directional and absolute trading detection
5.5 A theoretical approach
5.6 Concluding remarks
6 Conclusion and outlooks
A Wisdom of the institutional crowd – Supporting information
A.1 Determination of the crossover point n
A.2 Asset selection: a model
A.2.1 Asset selection in the small diversification region ni < n
A.2.2 Asset selection in the large diversification region ni n
A.2.2.1 Maximum investment ratio
A.3 Simulation of asset selection
Bibliography
References

Télécharger le rapport complet