They are mainly two kinds of imbalances that are important in the price formation
- the imbalance of liquidity providers (filling the orderbook in a LOB-driven market)
- the imbalance of liquidity consumers (sending marketable orders in a LOB-driven market).
It is quite obvious that these two imbalances are shaping the prices: if you do not change liquidity offering, the more buying consumers and the higher to price will go up (same for sellers and the price going down). And if you do not change the conuming flow, the liquidity provision imbalance will lead one queue to depletes before the other, driving the price up or down.
Moreover, you have to keep in mind that, since the best strategy to buy or sell a large quantity of shares or contracts is to split them (according to a balance between trading costs and opportunity costs), and because trading algorithms (and traders) mix limit and marketable orders in LOBs, the same "agent" usually contributes to both: a buying agent will provide liquidity at the bid and consume some at the ask during the same day: these effects are deeply mixed.
The point is how to to capture these imbalances with not too noisy estimators.
- the imbalance of liquidity can easily be captured by the imbalance between the bid side and the ask side (orderbook imbalance), on "large tick instruments", it is enough to focus on the first limits. See the second section of "Incorporating signals into optimal trading" by Neuman and L in Finance and Stochastics 23.2 (2019): 275-311 for illustrations.
- the imbalance of liquidity consumption can be captured by a difference between the sides of recent trades, and it is what propagator models are capturing.
The intrication of both imbalances are captured by the Queue Reactive (QR) model, linking the flows (or liquidity providing orders, but also of liquidity consuming orders) with the state of the orderbook.
The Order Flow Imbalance being the infinitesimal increment of liquidity provision, it is linked with the orderbook imbalance (ie if you sum the OFI between $t_0$ and $t_1$ you obtain the difference between the quantities on the two orberbooks: $OB_1-OB_0$). It is clearly capturing a part of the formation process. Moreover, because of the relation $OFI(t_0\rightarrow t_1)=OB_1-OB_0$, it can be used to describe the trajectory of the orderbook in a bid-ask quadrant: put the quantity at the bid on the x-axis and the quantity at the ask on the y-axis:
- the state of the orderbook is a point $X=(Q^A,Q^B)$ of this quadrant
- the moves of this point are exactly specified by the OFI, i.e.: $dX=OFI\,dt$.
This view has some limitation: it has been shown by the QR Model that the first limit is not enough, and that the "jumps" consecutive to the full depletion of one of the two queues (first bid or first ask), is conditioned by the shape of the orderbook.
In the second edition of Market Microstructure in Practice (L and Laruelle, 2018) the chapter "The Price Formation Process and Orderbooks Dynamics" describes all this in detail.