Linear or Nonlinear prediction model ? If people believe that price and volume contain all public information, that means we shall be able to forecast future market movement based on price and volume. Then, the question comes to which prediction models to use. Do you think if linear prediction model will be able to forecast market movement, or do we have to go to some non-linear prediction model ? What do you think ?

Huh? You just are leaping a huge chasm of logic there. Maybe price approximates closely what people think the value is but there is no reason to believe that you can predict the price without assessing what EVERYBODY thinks. Good luck with that s&^t btw... People have gotten PhD's proving that you can't predict future price from a data stream. That tells the prudent mathematicians among us that the best you can hope for is a partial solution and you still won't be able to account for "asynchronous" [my choice of terminology] events like a big order appearing out of nowhere.... Fractals

I do not get your points. 1st: If price reflects people's thinking, what do you mean "there is no reason to believe that you can predict the price without assessing what EVERYBODY thinks. " 2nd: a partial solution is good enough. As long as we have a little edge to beat market, we can use tools to amplify the edge, right ?

You said that price and volume reflect what all people think something is worth. To predict what they are thinking next is a big job. If you want to swim upstream for the rest of your days fighting mathematical realities you can go against the proofs that you cannot predict a future price from a data stream. Sure, PhD's don't for the most part make money trading, in fact they can lose so spectacularly that they can almost take down the whole financial system ala LTCM. They revel in the glory of intellectual stuff. Nonetheless they have proven that you can't predict a price from a price series. Volatility is a little predictive of volatility and volume can tell you when an "asynchronous" event is occurring and that is about all you have to work with in a mathematical sense that I'm aware of.

My understanding is that the verdict is in, mathematically you cannot predict the price of the next data point. Fine, I say, look for some kind of a partial solution and build on it or do an empirical thingy like a poster on this thread linked to. The channels represent volatility and the volume thingy shows when the market is making a turn. Personally I think that a lot better solution exists than that empirical thing but it's one way to go.