Here's a plot of a correlation function between log-returns and "order flow" (trade sign * volume):
[img]/User%20Files/10010/order_flow.png[/img]
This is for AAPL and I downsampled the data to 1 sec for simplicity.
So if the returns are denoted as r_t and order flow as o_t, then the correlation function is (x-axis is variable s in seconds)
[img]/User%20Files/10010/Latex-Equation-12267.gif[/img]
(it's a function of the difference s only since both series are stationary).
The correlation between past order flow and future returns is not very strong... there's also an interesting asymmetry in that the right hand side decays faster than the left hand side.
Does this mean that order flow is in a way a used up resource as a price movement indicator? Too many sharks feeding on one whale?
PS: the NP equation editor/TeX is not very good... why not use MathJax??