Summary
CrossAssetOrderFlowImbalance estimates the rolling linear impact of a peer order-flow
imbalance series on the own asset return: \(\beta\), fitted impact \(\beta\cdot\mathrm{peer\_ofi}\),
and residual return. It is a Cont-style cross-impact feature for multi-name flow.
Update API
import rtta
ind = rtta.CrossAssetOrderFlowImbalance(window=50, fillna=True)
result = ind.update(own_return, peer_ofi)
# result.beta, result.impact, result.residual, result.peer_ofi
advance(...) updates state without returning a result. Batch accepts aligned
own_return and peer_ofi arrays.
Theory Of Operation
Given pairs \((r_t, f_t)\) where \(r_t\) is the own return and \(f_t\) is peer OFI (or any
peer flow feature), the class maintains rolling sums of \(r\), \(f\), \(r^2\), \(f^2\),
and \(rf\) over a window of length \(W\). Beta is the OLS slope of \(r\) on \(f\)
(no intercept term in the moment form used by RollingPairStats::beta):
Cross-impact and residual then follow the linear model \(r \approx \beta f\).
Recurrence
After each push of \((x_t, y_t) = (r_t, f_t)\) into a window of capacity \(W\), with running sums \(S_x, S_y, S_{x^2}, S_{y^2}, S_{xy}\) and \(n = |W_t|\):
If fillna=False and the window is not full, \(\beta\), impact, and residual are NaN
while peer_ofi still echoes \(f_t\).
Implementation Notes
The recurrence is implemented in src/rtta/indicator.cpp in
class CrossAssetOrderFlowImbalance using RollingPairStats (x = own_return,
y = peer_ofi). Window length is at least 2.