What is needed at this point is a demonstration of the results for a price - volume system (as opposed to use of volume as an adjunct to other indicators as many of us use) that gets the discussion beyond that of an academic exercise.
I have a lot of sympathy for the Socratic idea that the full understanding of volume requires extensive observation of a market (although I would also expect there to be some core principles of volume trading common to most liquid markets). But here is a dilemma:
either Socrates is saying there is a system to uncover, but it has lots of complexity and many parameters to estimate, in which case he is just telling us it is a hard problem;
or he is is saying that neither can you systematise it, nor can you test your attempt at systematising, in which case he is then telling us that this is a theory with no empirical content.
My reading of Socrates is that he has embraced the first horn of the dilemma. Me too. And possibly the problem is so hard that there is no practical way of systematically generating the signals we wish to test. But the proponents of volume are surely claiming that a satisfactory study of the volume footprints (of a chosen instrument) does lead to practical understanding (=ability to generate value in trading), so probably they do not believe it is really "too hard" in this sense.
Skimbleshanks's claim that volume can be used for precise timing but cannot be subjected to back-testing is either just wrong or requires a more subtle understanding than I have given it so far. On the face of it, any numerical indicator that can be acted upon leaves a trace, such that inspection of the past would show hypothetically how it would have traded. Skimbleshanks cannot be denying this. Perhaps the soft area for Skimbleshanks is in the idea of back-testing, and the sense that back-testing leads to "curve fitting", and hence is of questionable value?
My answer to this is forward testing; set up the proposed signal on software that generates real time alarms, or actually trades the signals without personal intervention. I have done both for limited periods (as well as back-testing). Of course, forward testing has the disadvantage that it is difficult to exhibit equity curves as Culion demands, without allowing time for an adequate trading history to develop.
There is also a paradox, that when you put together complex systems the performance of the whole is not a straightforward sum of the parts, particularly where the elements come into play in different sequences, under a branching logic. So, if the algorithm says to pay attention first to the strength of the trend, and only if trend is weak take contrary signal from volume, the latter will have a higher success rate than if there is no trend filter. Indeed, the combination may have a positive expectancy even when neither component is positive on its own.
As a straw man, the attached thumbnail from yesterday' s S&P shows the volume based signals from Tom Williams's
VSA/Tradeguider. The green ones below the price bars are indications to close shorts or open long positions, while the red ones above the bar indicate the opposite. They identify known combinations of volume and price movement where the probability of a price reaction is reasonable. (For instance, the red triangle at 16.00 is noting that the mini rally had no volume and should be sold). The colour of the bars indicates the system's view of the prevailing trend. However, the indicators are not themselves sensitive to the trend, and for this reason users of the software are invited to make their own judgement about contextual factors. Of the nine indicators to be seen on the chart, six propose a reversal of trend, and following them against the trend would would probably have cost you money.
The question (of which this is just
an example, not
the example) is whether value could be added to this by systematically - and testably - integrating the VSA type indicators with the trend, in a way that the Tradeguider software does not itself attempt. But I take it that if somebody were to carry out that project, it would be the sort of thing that Culion is looking for.
Of course, that does not rule out the existence of subtler and more powerful approaches to the use of volume, but the same questions would arise. What Trader 333 says about learning seems to me to be an example of this, not a counter example; what a human being can learn (in a finite arena like stock or futures trading by price and volume) is surely something that can be approximated by models, else it could not be learnt either!
