Welcome - 欢迎 - Willkommen

Andreas Joseph
 
PhD 

Research Economist 
Bank of England

Contact: 
andreas.ch.joseph [at] gmail.com






Research Interests:

Thesis title:

Network-based Data-driven 
Approaches to the Analysis of
International Trade and Finance



A short CV of mine can be found here.


Big data and complex networks 
are two of the buzzwords of our time.

One the one hand, proliferation and advancement of technology 
allow us to collect and process ever growing amounts of data on a 
scale unprecedented in history.
One the other hand, the awareness of interconnectedness of a 
globalized world is rising. The flow of information, people, goods, 
money and social contacts leads to the formation of highly-complex 
- by themselves interconnected - network structures in the form of 
social networks, trade webs, transportation networks, the internet, etc.

Motivated by these facts, this web-page presents statistical tools for 
the analysis of complex data structures in general, while the leading 
paradigm is the analysis complex networks.

As a data structure, complex networks have the advantage of higher-order correlations between individual actors being automatically built in.
In addition, recent years saw an inflationary growing interest in the
field of network study which leaded to the development of a great
variety of theoretical tools and measures, most notably a zoo of node and edge centrality measures.

In a given situation, a hand full of such indices might be of interest
simultaneously, while it is a non-trivial problem to treat several 
- probably correlated - measures parallelly.

The projects presented here are dedicated to tackle this problem by
introducing several statistical tools, and demonstrating there application to real-world networks. 

See Projects for more information.


Some content may be out of date, sorry.

Summary Posters:

Snapshots:

Global Balance of Payments Network:
https://sites.google.com/site/datatoolx/home/BoP.png
International trade and investment activities of different countries are observed to be non-trivially interconnected and interdependent.
In this sense, they are expected to describe/forecast each other. We present an iterative multiple linear regression fit algorithm, which establishes phenomenological fit-relations (links) between individual country statistics (nodes) for a representative group of 60 countries and 8 indicators: In-coming and out-going cross-border trade, equity- and debt securities investment and foreign direct investment (FDI) during 2001-2011. 
The visualisation of the obtained network model, which combines propertiesfor description and forecasting from network science  and statistical modelling, shows that there are many indicators of small size (horizontal axis), which point to large indicators, resulting in a high tracking centrality (vertical axis). Such indicators may be taken as economic thermometers, pointing to larger changes in the global economic environment. In addition, this network model is able to forecast next-year values for the majority of the approximately 400 indicators with an error of less than 10% for an out-of-sample test.

The G-20 Trade Web as of 2011:
https://sites.google.com/site/datatoolx/home/g20_wtw_2011.png
The G-20 major economies account for over 80% of world-GDP 
and world-trade. In the above visualisation, the mutual trade flows 
are depicted, where the thickness and size of  arrow heads are 
proportional to the amount of trade between two countries. 
Node size is proportional to the log-GDP of that country.

The Gate-Keeping-Potential as a Proxy for GDP-Growth:
https://sites.google.com/site/datatoolx/home/g20_GKP_gdp_gro_corr.png?attredirects=0
The local gate-keeping-potential (GKP) is a topology-based network 
measure, which quantifies to which extent a node (country) is able 
to funnel trade flows through itself, which is interpreted as a measure 
of control within the network. 
For most countries, GKP exhibits a positive temporal correlation (1991-2011) with annual GDP-growth - contrary to more standard indicators such as the total value of imports or exports.

Network Warning Indicators for Financial Crises:
(click to enlarge)
https://sites.google.com/site/datatoolx/home/nlsmm_fits.png?attredirects=0
The edge density (red line) of the global debt securities investment network can be used as the input variable for a simple phenomenological model to describe the proliferation of certain financial derivative products; here credit default swaps (CDS, left), equity-linked derivatives (ELD, middle) and "hidden" un-allocated derivatives from the BIS semiannual derivatives statistic.
Setting warning thresholds proportional to world-GDP (gray dashed lines) within a comfortably large range, synchronous warning signals (red dashed lines) are generated before the onset of the global financial crisis 2008.

More information under:
Scientific Reports: nature.com/srep03991

Network Genetic Fingerprint 
of China and the US in the world trade web (WTW) from 1970-2010:
Bar heights show the individual and combined contributions
of both countries to their relative integration (centrality)
measured in terms of standardized graph measures.

See Software for a ready-to-use Python implementation.


Measure Standardization for node centrality inside the
world migration web (WMW):
Convergence of the composite node centrality distribution 
of standardized measures to a universal standard Normal 
distribution.

Well-defined inter-Network Node Comparison:
Comparison of evolution of composite node centrality scores
after measure standardization. In the WTW, one observes
a clear convergence of the centrality scores of indutrialised 
countries and the BRIC-block of emerging economies.
This can be interpreted as an act of global balancing or
globalization.

More information under: 
ScienceDirect (Physica D, subscription needed)