I have a set of financial "positionbook" data that shows the percentage of traders who are holding "long positions" versus "short positions" in a financial instrument at specified times during the day. For example, initially at time t1 the "long positions" of traders is say 80% and for "short positions" it is 20%. Trading then occurs and twenty minutes later at time t2 the percentages have changed such that "long positions" is now say 85% and "short positions" is 15%. These total percentage values can additionally be broken down into smaller percentage values according to price levels and, dependent on these levels, the percentage values can be broken out into additional categories thus:
- price levels (and associated percentage values) at which the real, absolute number of traders holding a position can only decrease (the majority of the levels)
- price levels (usually about 2-4 levels) at which the percentage values reflect the net change resulting from traders both acquiring/entering new positions and closing out/exiting existing positions and therefore the real, absolute number of traders holding a position can increase or decrease
Additional information available is a rough proxy for the combined number of all executed trades over this twenty minute period, e.g. a tick volume total of say 500.
Ideally what I would like to calculate from this limited information is to infer the changes in the underlying absolute number of traders' positions and then apportion the 500 tick volume into buying and selling pressure categories as evidenced by the changes in the "long positions" and "short positions" percentages. Even getting to an accurate ratio between the two would be useful.
Note that this is NOT order book data and so there are no concerns about whether a trade is possible or not; this "positionbook" data shows the positions of traders based on trades that have actually occurred and result in actual traders' positions in the market.
Is this even possible with the limited amount of information available? Suggestions as to models/algorithms/numerical optimisations to try or references to online resources are welcome.
Note: a "long" is a position in the financial instrument that will profit if the price goes up and a "short" will profit if it goes down.