Static web content is hosted on GitHub Pages - (http://nimmaj.github.io/sentiment-trading/) note that the demo isn't static so doesn't do much unless you run the app...

You can see a short screen cast of the presentation here.

Having won the Deloitte Gone Hacking Hackathon last November, there's basically no chance of us getting any prize this time, so we tipped up to have some fun!

Our plan was to investigate whether we could use social media, and any media, to give early indication of job changes and whether we could use that to then trade.

Our strategy was threefold:

• Investigate whether there was any statistical link between the step functions on non farm payroll days and the difference between the expected and actual numbers (Mark)
• Scan social media, and media, for job related events/sentiment that we could grade with a confidence and directionality to drive position taking events. (Yongzhi and Richard)
• Build a platform that could take live events and trade, but which could also be used to store and run historic back testing so that we could tune our data gathering and trading stragegies (Ben)

We used various technologies:

• Vert.x - we love vert.x because it's lightweight, polyglot and has an eventbus; all of which helps us to componentise work which helps us not step on each other's toes during the very short hack
• R - mark made extensive use of R to understand the times around the non-farm payroll events
• OANDA - awesome streaming rates and fx trading
• Bootstrap - a huge time saver when knocking up websites
• Streaming graphs from Smoothie.js - I'm a huge fan of the awesome Joe Walnes and Drew Noakes - thanks for these fun components!
• Python - Richard has this strange fascination with his python - we're hoping counselling might help... ;-)
• NLTK - Standing on the broad shoudlers of all the great work that has gone into the [Natural Language Toolkit] (http://www.nltk.org/)
• The Dovetail - one of the finest purveyors of ludicrously strong Belgian beer in London!

Huge thanks to the Deloitte Digital/Capital Markets GoneHacking team who have once again made this a fantastic, and fun, event.

You need a file in mods/oandastream1.0/conf called secret.json of the form:

{
  "accountId": <yourAccountId>,
  "apiKey": "<yourApiKey"
}

Run with:

./run.sh

run using

mongod --dbfile=blah

use the mongo shell to access the data:

use sentiment_db
show collections
db.event.find()
db.event.remove( {} )

db.event.aggregate(
   { $group: { _id: { $week: "$timestamp"},
               click: { $sum: 1 } } }
   )

For backtesting and historic event parsing you need Python2.7

python /parser/parsing.py