Paul D. McNelis, "Neural Networks in Finance: Gaining Predictive Edge in the Market"
Academic Press | ISBN 0124859674 | 2004 Year | PDF | 1,7 Mb | 256 Pages
I've done some work in this area during my post grad study. There are two problems
I can see with neural network.
1. The number of nodes you will need is up to you. But more nodes means more
computation is required. They do go up substantially if you fully implemented the
entire neural network properly. Statical analysis takes a lot of time.
2. You need large amount of data to properly train your system. Even that, how
good is the system is still questionable.
Almost 15 years ago I implemented a single node neural netowk according to the
first neural network paper. On a 8086 PC, I trained it for three days and needless
to say, I never finish.
Please don't get discourage though, you maybe smart enough to do it.
In fact, I understand about the architecture of the neural network is up to me. The more node I use, will increase the training time. I've implemented a fingerprint recognition system using neural network, the training takes me 7 hours using Pentium 4.
I wish to know, if I want to implement a share price forecasting system using neural network, which algorithm is suitable and what data I should use as the input.
Is the share price predictable on a short term basis? Many professional account managers use their own mathematical model to say when to buy or sell. Your program will have to do those calculations to predict the number of sellers vs the number of buyers. Then there is the effects of company profits on share price. You will have to predict the profits of the companies without having access to their accounts.
What inputs shall we use to predict share prices? Do we have to include the company's condition or maybe the market condition? How if i only use common data, such as higgest price, lowest price, opening price, closing price, and index, from the past year maybe?
Yes, this is for short term basis, maybe 10 days or 1 month. Actually this program helps to predict the price in term of historical pricing. After the prediction, then we have to study the company performance manually. So, I hope can develop a program for this purpose. Any idea?
I wish to know if just put in past day price and volume, is it enough? How to put this in? Take the average or put the figure in directly?
maybe u should know the average 1st, there's a lot of math function u will use here.. i just wonder what input should we use in this case? are these such as highest, lowest, open, or close price, enough?
Actually, I have friend who is a Trader have thought of it. But later, he did not complete the idea project due to some reason.
Anyway, I think U need to include some parameters that can recognize those patterns like H&S. It is very important. Maybe, it is not very critical for applying the program into the KLSE market.
But if U are entering some volitile markets like HangSeng, Nikkei and Dow. Then, U will have big trouble.