Hakia's SenseNews: Can it really tell you when to buy and sell stocks?

There is a always a big difference between an exciting technology and an exciting business. And understanding those differences is the diffference between success and failure. Hakia always had an exciting technology if you ever cared to go to their web site and tried to evaluate their semantic search results. Unfortunately, they never had a succesful brand name like Google, Yahoo etc. which is probably true for most of the startups in the semantic search space. As a result, they could never attract too many visitors to their website. What got my attention is their recent announcement about Sensenews which is a service to make buy and sell recommendations for any stock. Sensenews reads news sources (over 30,000 news sources), blogs (over 1 million), Twitter and performs an advanced computation to make buy/sell recommendations to you. 

Stock movement is always correlated with news - good or bad. An example in the fall of 2008 shows how unexpected news can impact prices dramatically. At 1:37 AM, September 7, 2008, Google's newsbots picked up 2002 story about United Airlines filing for bankruptcy. The story had a cascading affect and it was picked up by a person, who obviously failed to notice the date on the story, and put it on Bloomberg. The stock droped 76 percent in six minutes with a huge spike in volume.

The issue with news is that it is not mathematical in nature. It is very hard to make machines understand news. Various semantic technology, text extraction, natural langauage and statistical techniques are used to maker sense out of vast amount of news for a particular stock, currency or commodity. Hakia is not the first one who is trying this as:

• Monitor 110, funded with 20 million dollars, was one of the first high profile failure in the capital markets and Hedge fund space. They had signed up almost hundred trial clients. They closed the shop in July 2008.
• Relegance (a company out of Israel), one of the last acquistions (between 50 million to 100 million US dollars) of AOL, was also in this space. AOL is often accused of not leveraging Relegence’s technology effectively. But then AOL is accused of doing lot of things not in a right way in last few years. Relegance's technology is being used for AOL Money & Finance and WalletPop.
• BNP Paribas developed their own indicator that reads the news looking for words that imply strength or weakness, and decides how to trade depending upon the number of times various words are used.
• German exchange Deutsche Borse acquired Need to Know News, a Washington D.C.-based provider of machine-readable news for automated trading engines.
• Reuter's NewsScope and Sentiment Analysis product is designed for use in algorithmic trading, where computer programs make all of the trading decisions including the entry and exit of any trades. More than half of the trading is now automated, much more for US stocks.
• Stocktwits - The service takes financial related data from tweets and structures it by stock, user, reputation, etc. 

This is a very hard and known problem which is becoming more painful because of the information overload. The whole idea is to:

• Exploit market inefficiencies by showing the unrealized value of a stock
• Manage event risk by reacting to breaking developments

Probably, not even 2% of the firms in the capital markets are doing this. There are hardly known successful cases - there is always a secrecy aspect also. But I don't know any vendor who has truly succeeded in making a successful business out of it. Semantic technologies are sometimes viewed as a technology which is looking for a business problem to solve but this case is different. At some level, the business problem is defined (yes, the requirements of every trading desk or traders in different firms will vary), but the technological limitations are exposed by its inability to semantically filter the right content, duplicate removals, understanding the right sentiment, categorizing it correctly, getting rid of marginal news, irrelevant stories and many other things. I also think that it is very hard to understand the social media manipulations as it is an opportunity for someone to make lots of money by fooling so called intelligent programs.

What is interesting in the case of Hakia is that they have gone a step ahead by building a recommendation model on top of semantic filtering of the content. This is really bold and interesting! Their delivery model is subscription/consumer based which is different from most of the ones who have tried this before or trying now as they were/are more enterprise focused with the exception of stocktwits. I do think that they should have offered thirty day free trial period for this service from marketing perspective. It is not clear what level of fundamental and technical analysis is done. I also don't understand the profile of the trader they are targetting. It also doesn't seem to be realtime in nature and won't work correctly for stock splits. But if they can deliver based on what they are claiming then it is commendable. Only time will tell whether they will be successful in this new business where no company has truly succeeded yet. But one thing is for sure that they have understood that future of semantic search - that it is very much related to being some kind of content clearing house i.e.  clean, curate, collate and categorize content.