That's what Reuters is doing. The big question is how to extract information from the meta tags and make semantic analysis. Neural Nets are not going to help in this case.
I'm not sure. The point is also what you need. Let's make an example. Let's suppose that someone (Israel or US) bomb Iran's nuclear facilities. Now, to really make something out of it a computer should know: what Israel is, what Iran is, in what region they are, what the region means for oil, what bombing means, what nuclear facilities are, and some other bits of information. But let's suppose that you can just make a computer understand that something bad happened in the middle east, and that there is danger of escalation. A neural network could learn, from similar events in the past years, that every time something bad happens in the middle east the oil price increases. A neural network should be able to do that. Of course I'm making this up, I have no evidence that it would work, and I'm not sure myself. It's just an idea.
I'm into AI and be assured it doesn't work that way. In the 80s one had Expert Systems and it just didn't work out. What can be used are techniques from computerlinguistiks (latent semantic analysis for instance). I heavily doubt that this is "mainstream", because the methods only exist in academia are generally not applied to trading. Perhaps in 20 years things will change. Anyway, let's assume the system knows something bad has happened. What are you going to make out of it, that is the real question. In the end political events would be the last events I would use for a system. I would use corporate earnings and macro-newsflow.
You'll note, I didn't say I agreed with the consensus. Indeed, I explicitly state semantic news analysis is one of the goals of the system I am developing: http://www.elitetrader.com/vb/showthread.php?s=&postid=1280530#post1280530 Earlier ET thread on the topic of news algorithms: http://www.elitetrader.com/vb/showthread.php?s=&threadid=63416 That thread is nearly a year old. These ideas that alledgedly nobody is using are not new. The fact that they have deemed fit to provide machine consumable tags in the first place is an indication of the state of affairs.
Well, expert systems are not neural nets. There's no reason why neural nets could not be leveraged at some stage in the decision-making pipeline but that is probably a separate matter. Techniques such as probablistic latent semantic analysis are in use in diverse areas. Companies with large amounts of information to get to grips with can use commercial products such as: http://www.recommind.com/ Or, for summarizing and sentiment analysis: http://www.corporasoftware.com/ Not to mention the $59 summarizer from Copernic: http://www.copernic.com/en/products/summarizer/download.html Then there are open source tools such as: http://knowledgesearch.org/ I hope some of the links above demonstrate that these methods are not only in academia. Seriously, I'd like to know where you get your information from If you are talking about Joe public sitting in mom's basement in underpants trading on laptop having access to this technology then I would agree that it hasn't reached that level. However, even a single technologically savvy independent trader with minimal $ and an ability to use Google can harness some of these techniques...and I know there are those that do. Either way, I wish you luck with your adventures.
Of course you are right with the average joe and thanks for the links. Actually I was not talking only about news interpretation. My question was in fact where do I get more information on these topics? Not only information, but how can I apply these technologies. Note I'm not working in an environment where I don't have access to some things. But on the other hand most of these topics are hidden anyway. Just google for trading systems and you will get 99.9% crap (mostly TA), so it's obviously not so easy as some of you imply. Also just having access to some libaries doesn't get you very far. You need the knowledge to integrate the different components into one sound system. There are some other fields I mentioned: behavioural finance (models) and agent based modeling, aswell as new algorithms for riskmanagement/moneymanagement and integrating technical and fundamental data.
For scheduled news, like non farms payroll, the semantic analysis may not be necessary. Just before the news you might submit two entry orders: a buy stop and a sell stop, or a buy and sell that gets one side cancelled when the direction is clear. I suspect that that is how people who trade such news do a manual trade, if they're not just guessing and hoping or have advance information about the report. The advantage to automating it is that the machine can enter/exit more quickly. A problem, doing it manually or automatically, is that the initial reaction isn't always the final reaction.
@TraderMojo: I went through your thread. Certainly very interesting from the technical side. I'm thinking of your post where you summarize: Price(?) -> Blackbox -> Orders That's where my problem is. It's that technical analysis is a very weak foundation for ATS.
"technical analysis is a very weak foundation for ATS" All of technical analysis is a very weak foundation? Would any of it be useful? What would be a stronger foundation? I think a moving average crossover system could make money at some times and lose money at others. If I could encode some rules about when to go with the crossover and when to fade it, possibly based on some additional technical analysis, couldn't I come up with a decent system?
Of course many will oppose to my proposition. To explain what I mean it is best to review the concept of technical analysis. The theory can be in the west tracked down to writings of Charles Dow. The heart of the dow theory is actually in my view intermarket analysis. I think the best book on technical analysis is the famous one by Edwards/Magee. In this book they partly describe how price patterns evolve out of behaviour of market agents. Most people use TA the other way round. They say this is a trend and people behave according to the trend when it is actually the other way round. A chart pattern sometimes indicates certain behaviour, but the link is very crude.