"Chapeau" is all i can say to that ;-) If my average return is twice the stdev of the return, than I'm already in heaven. Not sure though how biomarkers relate to prepayment risk.... Oliver
Well... Google gives ZERO trading related hits... for "immunoligic trading" and "trading algorithm". Always a very bad sign. The webcast you keep pointing us to... Is a generic sales pitch/overview... Of why your firm MUST absolutely buy their products. I think you're just using fancy terms... For "expert systems" which adapt/evolve/learn in a VERY crude manner. People... just look at your life. How many "expert systems" are part of your life? Does a computer diagnose your illness? Does a computer fix your car? The answer is NO... Because "expert systems" have not fulfilled their promise. "Evolving bots" trolling financial markets that are mostly random systems... Are no more than a misguided pipedream... Or a Machiavellian sales pitch. rm+
Here's some interesting looking articles I downloaded in case anyone is interested. I haven't gotten a chance to even skim them yet. (I plan to read them next week during a flight.) Immunity by Design:an Artificial Immune System An Artificial Immune System Aproach to Semantic Document Classification Feature Extraction CNN Algorithms for Artificial Immune Systems Design of an Artificial Immune System as a Novel Anomoly Detector for Combating Fraud in the Retail Sector Here are also some books on Amazon that I just ordered: Artificial Immune Systems: A New Computational Intelligence Aproach Artificial Immune Systems and their Applications
Differentiating between "self" and "non-self" properties of price behaviour... In a changing environment that would presuppose defining a rate of change and an algorithm exactly measuring how "self" is morphing to "non-self" and viceversa. Given the fact, that the rate of change of the environment is not known and we're probably dealing with a stochastic process, it would be pretty damn hard to construct a precise algorithm for a necessary process of "forgetting" or "re-learning". The possible effect might be an "immunological reaction". I don't think successful fraud detection prepares you to even imagine the complexity of the system you're dealing with. Another aspect would be competition reverse-engineering your algorithms to fade them in a stochastic, automated manner, triggering risk-management. Hm, well that sounds profitable...
Let's start with the assumption that the state of artificial intelligence right now (as envisioned in the 50's) truely sucks. All the human-esque capabilities that early computer scientists thought would be easy (vision, locomotion (bi-ped), thinking (whatever that means), learning, etc) have turned out to be very difficult. No disrespect to anyone else here, but if I hear another "Deep Blue" comparison to human capabilities.... I will not be responsible for my actions. Deep Blue was a complex super computer using advanced custom built software. Not to say it was not an accomplishment - it was - it just was not an "Artificial Intelligence" (or machine learning or adaptive, etc) accomplishment. It basically just searches a bounded set of possibilities - brute force (here is my rant on brute force computing). Try that with Go. Not to say we can't use AI algorithms to help us trade - heck, that is what I love doing - but the bottom line is that the human is still going to be an important part of the cycle for a very long time.
Let us know what you think GTG - I am curious to hear. When I read through this AIS stuff it just sounds like genetic algorithms and evolutionary computation, which have been around since the 60's. I can't get a sense of the real difference. Seems like a "buzzwordification" of an existing field.
What woudl you say if if I told you i developed a set of algorithms that take in news to trade which has a win loss ratio of > 11:1 over the last 24 months. Would you say that's unlikely?
No - I don't argue with success. If you are - excellent! It is just my experience that I have not come across anything out there like you are describing - it is just too difficult. I would certainly love to hear about your methods and results.
Well, 8:2 is superexcellent but I know it can be done, from my own experience. 11:1 is so far ahead of that , I cannot imagine it is possible. Especially if you seems to be implying that your method is fully mechanical. If you are saying that you can discretionally do it, I would not argue with that. Although ,I would be still skeptical till I saw it with my own eyes .