Using sentiment analysis to predict FOREX quotes Print

I worked on some tools for . That service has algorithm for sentiment analysis of business news. For each message there is list of Tags and sentiment (positive or negative) for each tag . Sentiment is calculated using Sentiment Analysis methods.

Using gathered data we tried to use Tags' sentiments for prediction of Forex quotes

We get interesting results. There are sets of Tags for each currency pair that can predict short term trend for 1-3 days.

For different time periods tags for prediction can be different. For example, tag "600 billion" had fine correlation with EUR/USD for period from  15 Oct to 15 Nov. Later this tag was not useful, because news related to "600 billion" appears less often now.

How prediction tool works.

Each day for currency pair (let it be EUR/USD) there is procedure of calculation of maximum correlation among all tags that appears more then 60 times for last month. Most correlated tags are selected and then there is alert when some of that tags appear in new messages. Using sentiments  of selected tags there is calculated averaged prediction for all selected tags during a day.

3 separate sets of tags are selected: predict for 1 ,2 and 3 days. Each set of tags gives prediction using sentiments from last 3 days.

This tool is in alpha testing now. I can't say how useful it will be. But this definitely can be some interesting tool for FOREX traders.