Clustering is one of our brain main ability. We use classification, patterns recognition on a daily basis.
We can create data trees by successive layers of classification. For large time series we need some help from an expert or an algorithm.
For instance, the clustering should provide a clue on Telecom Italia credit.
Is Telecom Italia credit a Telecommunication credit or a fallen angel credit or an Italian credit?
Clustering classifies, splits and extracts knowledge from data.
Clustering is not unique. It is a craftsman work.
1.1. Expert Tree
The Expert Tree aggregates the data according to our experts’ data analysis.
For instance, for financial time series, our experts are using the market microstructure: quotation place,
asset quality, industry sector to create an intuitive and performing Data Tree.
3.2. Machine Data Tree
The Machine Tree aggregates the data according to the joint evolution of the time
series using a method developed by Hellebore Capital Management using machine learning technics.
For CDS data, the calibration is achieved over the whole data sample. The Machine Tree is updated on a quarterly basis.
2. About the Grapple
A Grapple combines the relevant data, the appropriate data clustering and the most suitable graphic visualization tool.
The user gets in one click a powerful, intuitive and useful data insight.
A Grapple is a ready-to-use data analysis for everyone.
offers the catalogue of Grapples.
The Grapple are classified per Data Type and Data Tree with some preferred Grapples.
The My Grapples page allows you to save your favourite Grapples.
The page saves the 10 most recent Grapples you have accessed to.
3. About the Tokens
Using a token enables an unlimited consultation of a Grapple during one day on any terminal.
You have a Token account in your profile.
You need to have some credit to have access to the My Grapples section.
So, if you want to enjoy all the features provided by DataGrapple, please register.
It is free.
You cannot borrow or repo the tokens, but you get 15 free tokens once a week, every Friday evening.
You are not charged for using DataGrapple unless you want to use more than your free tokens a week.
If you want to purchase additional tokens, please contact the Customer Service.
The price is only €0.49 per token.
If you brought tokens, those will be only used once your free tokens are used.
The purchased tokens have no expiry date.
Using DataGrapple you are just charged for what you use and what you like.
Hellebore Capital Management is sponsoring the project for its clients.
Hellebore Capital Management can credit tokens on your account.
Please ask to your representative: email@example.com
4. About the Supported Environment
DataGrapple is working with all modern browsers, namely : Chrome, Firefox, Opera and Internet Explorer from version 9.
Datagrapple is designed to work on tablet (or larger) screens. Mobile phone users may experience a degraded experience.
5. About the Releases
DataGrapple has been released on January 29, 2014.
The first version includes a wide dataset on the credit market: 400 corporate credit default swaps,
the Markit® credit derivatives indices and the Markit® credit derivatives indices bases.
This version includes the CDS volumes reported by DTCC TIW.
The DataGrapple project should release new versions on an ongoing basis.
The project is focusing on combining data mining technologies with graphic representations.
February 3, 2014
New Grapples ranking the CDS as a function of their performance
Addition of the year to date (YTD) in the time frame selection.
June 3, 2014
New Grapples using a scatter plot for analyzing the CDS term structure and term structure evolution.
September 12, 2014
New design using HTML5 and D3.js. All Grapples design has been improved.
Addition of the Data Tree navigator on each Grapple to ease the drill down in the data tree.
Introduction of the GrappleStore and My Grapple pages to ease the access or the selection of the Grapples.
Creation of a free access blog. The blog leverages a Grapple with a comment on the data time series.
Capability to open DataGrapple to other time series.
November 27, 2014
The CDS list has been increased from 425 to 680 reference entities. DataGrapple includes the 500 most traded CDS according to DTCC and the on-the-run credit index members including Asian credit index.
Improvement of the alleged CDS levels algorithm. We have used our machine learning technology to infer missing data in incomplete historical time series of recently introduced CDS, like LOXAM, newly added in the iTraxx Crossover S22.
Introduction of a free access Tech Blog. The first post explains the method used to design our Machine Tree.
Mailing of the blog’s new entries to all subscribers on a daily basis.
March 26, 2015
The CDS list has been increased from 680 to 697 reference entities. DataGrapple includes the 500 most traded CDS over the last semester according to DTCC. In addition, DataGrapple includes the members of the credit indices from series 19 to the most recent series, the series 23 in Europe and Asia, the series 24 in the US.
The machine tree has been updated with the most recent market data. The clustering algorithm has been improved. DataGrapple’s machine tree is now the most stable machine tree using the past 9 years of data.
Alleged time series for new CDS have been updated using the past 6 months of market data.
September 30, 2015
The CDS list has been increased from 697 to 720 reference entities. DataGrapple includes the 500 most traded CDS over the last semester according to DTCC. In addition, DataGrapple includes the members of the credit indices from series 19 to the most recent series, the series 24 in Europe and Asia, the series 25 in the US.
The machine tree has been updated with the most recent market data.
Alleged time series for new CDS have been updated using the past 6 months of market data.
February 9, 2016
ABENGOA has been removed from the trees, because of its default on January 14th.
Probability of Default (PD) time series are now available in a new set of Grapples. The PD is computed with the credit reduced form model calibrated with the CDS prices and the market loss given default (LGD). For low 1Y CDS,
CDS ≈ PD * LGD
For close to default CDS,
CDS NPV ≈ LGD
A cluster's PD is the average PD of its members. Better than the CDS time series, the PD time series grapple with small and large CDS premium.
May 11, 2016
The CDS list has been increased from 720 to 732 reference entities. DataGrapple includes the 500 most traded CDS over the last semester according to DTCC. In addition, DataGrapple includes the members of the credit indices from series 19 to the most recent series, the series 25 in Europe and Asia, the series 26 in the US.
The machine trees (CDS and PD) have been recalculated adding the period from 20/09/2015 to the end of March 2016.
Alleged time series for new CDS have been updated using all data available as of the end of March.
DTV has been removed from the trees, because of its succession event by SBC in March.
BTU has been removed from the trees, because of its default in April.
A new kind of Grapple has been added, the sankey, which enables to compare trees. It represents flows between the nodes of two trees. A connector links nodes sharing the same components from a tree to another.
The size of the connector is proportional to the number of components it links.
October 10, 2016
The CDS list has been increased from 730 to 733 reference entities. DataGrapple includes the 500 most traded CDS over the last semester according to DTCC. In addition, DataGrapple includes the members of the credit indices from series 19 to the most recent series, the series 26 in Europe and Asia, the series 27 in the US.
The machine trees (CDS and PD) have been recalculated adding the period from 20/03/2016 to the end of September 2016.
Alleged time series for new CDS have been updated using all data available as of the end of September.
PORTEL and ISOLUX have been removed from the trees, because of their default events in July and August.
ARGENTINA, HNDA, MGP, PRUFIN, SSELN are the new entrants.
6. About Us
The team has designed many softwares from laser beam propagation simulation to
financial markets trading and risk management tools.
The team skills include modelling, statistical analysis,
database expertise, object oriented development and web applications.
The team excels in agile developments: testing and adjusting its developments
with its clients on a day to day basis. The team is currently developing software for
financial institutions but it is looking to apply its knowledge to other areas.
The data analysis has gained a renewed scientific interest over the last ten years due to the accumulation of large data samples
and the massive increase of computational power.
Many ready-to-use building blocks have been made available.
Bringing all the parts together in a user friendly application is quite a challenge.
Our ambition is to fill the gap between the recent technological developments and the decision maker’s curiosity.
The project is leveraging a very specific data, the CDS prices, but the DataGrapple project can tackle with any time series.
For any other information about the DataGrapple project or our company, you can contact us at firstname.lastname@example.org.