Credit Market Sentiment

In this exercise, you are tasked to build your own "Credit Market Sentiment Machine".

You are given a dataset comprised of 1,309 comments on the market written by portfolio

managers.

These market comments are highly specialized, full of jargon tied to credit derivatives, and contain their fair lot of abbreviations, cryptic tickers, and more or less funny puns directly translated from French. They also sometimes contain remarks on equities and FX markets which are less relevant to your task.

For each market comment, you are given its title, the content, and the date it was written. For example:

{

'title': 'Less Volume, Less Margin!

'content': 'Whiting Petroleum Corporation (WLL) saw its 5y spread

virtually double, i.e. widen by c290bp since July 31 (European) close. This

came after a very poor Q2 set of results and a revised guidance. They printed

an adjusted loss per share of -28c vs an expectation of +62c EPS, although

the range of expectation tracked by Bloomberg was rather wide (startingat

+12c) . But none of the predictors expected a loss. Realized price per unit of

both oil and gas were down in a double-digit fashion, compared to Q2-18. The

stock has lost c43% since July 31 (European) close. In 02 production was

affected by issues related to infrastructure constraints, which are expected

to remain until the end of the year. Not only the output but the price at

which the company sold gas has also been affected by these, on top of a

systemic oversupply environment. WLL therefore had to also reduce its full-

year output guidance, Which makes market participants wary that it is not

going to generate much cash flow before next year. Last month, WLL agreed to

sell some properties for $53m, that may weigh and also reduce full-year

output. Against this challenging backdrop, another alarming sign for markets

is that WLL is implementing an organizational restructuring that will reduce

its workforce by c33% (i.e. 254 staff incl. 94 executives) to save $50m on an

annual basis, after this plan will have incurred a (one-off) $8m charge. on

top of WLL total output being less than expected, its oil production more

specifically also declined in the mix. It seems that WLL's oil percentage

will be around 65-65.5% for the year, due to delays in new wells which happen

to have higher oil %.

date: '2019-08-051'

}

In this example, the sentiment is strongly negative for the credit of the firm.

Your goal is to build a model which can extract a credit sentiment, i.e. a score in [-1,1], for

any given credit market comment (in the same format as above), whereby:

-1 indicates extremely negative sentiment

O represents a neutral comment

+1 indicates extremely positive sentiment

The comments in the dataset are not provided with any sentiment score labels to begin with, so you will have to devise your own method(s) of annotating the dataset first if you plan to use supervised learning for building your model.

Please do all of your work in the included Jupyter notebook, being sure to describe your approach in detail, especially with regards to any modelling trade-offs. The person who will review your solution should also be able to easily download your model

from this Jupyter environment, and apply it on a testing dataset (same format and type of content) to extract a sentiment for each comment of the test set

market_comments is a file I can't read but has the exact paragraph in the example above when printing market_comments[-1]

the jupyter notebook has the following:

# Import libraries

import pickle import numpy as np import pandas as pd import matplotlib.pyplot as plt

# Read dataset

with open('market_comments', 'rb') as file: market_comments = pickle.load(file) print(market_comments[-1]) output:

{'title': 'Less Volume, Less Margin', 'content': 'Whiting Petroleum Corporation (WLL) saw its 5y spread virtually double, i.e. widen by c290bp since July 31 (European) close. This came after a very poor Q2 set of results and a revised guidance. They printed an adjusted loss per share of -28c vs an expectation of +62c EPS, although the range of expectation tracked by Bloomberg was rather wide (starting at +12c). But none of the predictors expected a loss. Realized price per unit of both oil and gas were down in a double-digit fashion, compared to Q2-18. The stock has lost c43% since July 31 (European) close. In Q2 production was affected by issues related to infrastructure constraints, which are expected to remain until the end of the year. Not only the output but the price at which the company sold gas has also been affected by these, on top of a systemic oversupply environment. WLL therefore had to also reduce its full-year output guidance. Which makes market participants wary that it is not going to generate much cash flow before next year. Last month, WLL agreed to sell some properties for $53m, that may weigh and also reduce full-year output. Against this challenging backdrop, another alarming sign for markets is that WLL is implementing an organizational restructuring that will reduce its workforce by c33% (i.e. 254 staff incl. 94 executives) to save $50m on an annual basis, after this plan will have incurred a (one-off) $8m charge. On top of WLL total output being less than expected, its oil production more specifically also declined in the mix. It seems that WLLs oil percentage will be around 65-65.5% for the year, due to delays in new wells which happen to have higher oil %. ', 'date': '2019-08-05'}