You may be right, it may contain the same information, one historical based, and one based on the current book. The question is which one contains more noise. I guess the goal of PA is not to assign SR or pivot lines -- those are obvious either from the historical data or order book. The goal is to differentiate the bounces from those SR lines as either short term (pullbacks) or long term (reversals), i.e., which ones are the real bounces (reversals), and which ones are the false ones (pullbacks along trend continuation) I guess traditionally people have not performed historical analysis of the order book (at least at the retail level, so that's why people gravitate to PA -- because of the ease of statistical analysis since historical price data is prevalent. If you have historical order book snapshots (e.g., every second), then one could do a statistical analysis on order books -- and since it contains more information, you will likely improve your predictions.
For excluding trades whose profitability is below the spread. Otherwise the algorithm finds too many patterns on short timeframes, f.i. 1-minute bars. Valid objection. But S&P500, NASDAQ, Oil, Gold gave similar negative test results as Forex. The main problem of the test is that it's limited to isolated 3-bar price patterns, and thus one should indeed not arrive at a general conclusion. However, 3-bar patterns are often recommended in trading literature for generating trade signals, and also a certain commercial software is based on them. So I think a test that reveals no predictive power of such patterns, at least under the tested circumstances, has some informative value.
Again I have to say that you not only considered a limited case of patterns, but also a very limited case of 3 bar patterns. I also know of no commercial software that is based on 3 bar patterns. Therefore I must conclude that we live on parallel universes maybe. In my universe patterns can have arbitrary number of bars, including skipping irrelevant bars; there is no commercial software that uses 3 bar patterns, etc. You seem to live in a very restrictive parallel universe of limited possibilities.
can't one just automate the ideas in this book: http://www.amazon.com/Profit-Prices...5131/ref=sr_1_1?ie=UTF8&qid=1332721446&sr=8-1
Why don't you do it and tell us your results? There is free software that can automate any technical trade method.
The orden book shows price action. It's true that it has noise. But when you come across a serious support or resistance it shows clearly (as a stupidly large size at ##.50 or ##.00) on the L2. This allows you to get in the move before it happens by looking at how size behaves. Eg when the big 5x5319sized offer at .99x.00 moves to 135x1500 and then to 300x250 the probability of it going bid ##.00 is larger than it going. Is not the size per size that matters its how it's changes. f(S)' Same can be said for price. How does it move coming close to the S/R level? All that. Can Be observed on the L2.
Thanks for this, this is really good info / motivation. I have been studying the order book lately and it does seem like it will lead to a viable strategy.
This gets a little bit harder with undisclosed and dark liquidity. If I am posting large, I'm not gonna do it in a lit venue. And, as for getting in before the move, what about order priority? Can't front run a block without paying up a tick. Issues like that. Swings and roundabouts + an expensive data feed.
Trying to understand what price will do from the orders in the book is like trying to guess which poker player will win from the size of the bets. Only historically significant patterns can provide some indication of future price action.
Reserve orders are a problem when reading the L2 to spot S/R. Dark liq not so much since it doesn't stoP the price from moving. When I talk about getting in early it is by paying the spread and lifting offers once the L2 shows that the prob is on the right sideâ¦