DISCLAIMER: THIS IS WORK IN PROGRESS. THERE ARE UNFINISHED PARTS, GAPS AND PROBABLY ERRORS!

KID for PRIIPs

In the PRIIPs (Packaged Retail and Insurance-based Investment Products) regulation of the European Union (No. 1286/2014) the KID (Key Information Document) is defined as a measure to improve investor protection. The document specified in the regulation consists of at most three pages presenting the details of a financial product, a risk indicator, different scenarios for the behavior of the product over time, as well a disclosure of the costs inherent to the product. In the annexes the exact form of the document and the calculation methodologies are specified.

Architecture

Calculations / Simulations

We implement the calculations/simulations for Category 2 and 3 KIDs. These are based on the following methodologies, resp. algorithms:

• Cornish Fisher expansion
• Monte-Carlo simulations

The GPU-based Monte-Carlo simulation framework is implemented in the programming language Futhark

The implementation is done the repository kid-annexes.

Payoff Language

A generic, certified payoff language has been proposed in the following paper:

Danil Annenkov and Martin Elsman. Certified Compilation of Financial Contracts. In Proceedings of the 20th International Symposium on Principles and Practice of Declarative Programming (PPDP’18). Frankfurt am Main, Germany. September 2018. PDF

The proposed language allows to generate payoff functions in Futhark that run on the GPU-based simulation engine for the KID calculations.

Historical data

Historical data is loaded from Quandl. In order to use the Quandl API without restrictions, an API key has to be provided.

Document generation

The documents are generated using LaTeX that are then compiled to a PDF document.

Building the project

Compile

Futhark

The dependency to kid-annexes needs to be established as follows:

$ cd src/fut
$ make pkgs
$ make

Haskell

$ cabal build

Run the application server

The application server is started as follows. Optionally provide the Quandl API key on the command-line:

$ cabal run kid-exe -- -k <quandl-key>

Sample request:

$ curl -f -X POST --data '@test/example/contract.json' -H 'Accept: application/pdf' -H 'Content-type: application/json' -o kid.pdf http://localhost:8081/kid?lang=EN

Note: pass curl -f (--fail). On an error the server responds with HTTP 400 and a plain-text message describing what went wrong (also logged to the server's stderr). Without -f, curl -o kid.pdf would happily save that error text into kid.pdf, producing what looks like an empty/corrupt PDF.

Historical-data backend

The source of historical data is pluggable, selected with --backend:

$ cabal run kid-exe -- --backend quandl -k <quandl-key>   # default
$ cabal run kid-exe -- --backend file

• quandl — fetches from the Quandl API. Underlyings use QuandlId instrument ids.

• file — reads from local CSV files. Underlyings use FileId instrument ids, each naming a CSV file and the value column to read:

  "underlying_instrument_id": {
    "tag": "FileId",
    "file_path": "test/example/data/underlying.csv",
    "file_column": "Price"
  }

  The CSV has a header row whose first column is an ISO date (YYYY-MM-DD) followed by named value columns; rows outside the historical window are ignored:

  Date,Price
  2013-11-01,100.00
  2013-11-04,100.31
  

  A runnable example is provided — test/example/contract-file.json together with test/example/data/underlying.csv:

  $ cabal run kid-exe -- --backend file
  $ curl -f -X POST --data '@test/example/contract-file.json' -H 'Accept: application/pdf' -H 'Content-type: application/json' -o kid.pdf http://localhost:8081/kid?lang=EN