Summary
WeightedMultiPeerOrderFlowImbalance is RTTA's streaming basket peer OFI with
explicit peer weights: a weighted mean of peer OFIs each tick, then a rolling
beta of own return on that peer mean (impact and residual).
Update API
# peer_ofis and weights shape (n_peers,)
result = rtta.WeightedMultiPeerOrderFlowImbalance(window=50).update(
own_return, peer_ofis, weights
)
# result.beta, result.impact, result.residual, result.peer_mean
# batch: own_return (T,), peer_ofis (T, P), weights (T, P)
batch = ind.batch(own_return, peer_ofis, weights)
Equal weights recover the peer mean of MultiPeerOrderFlowImbalance.
Non-positive or non-finite weights are skipped; if every weight is invalid, the
implementation falls back to equal weighting.
Theory Of Operation
Equal-weight basket OFI treats every peer name the same. Weighted multi-peer OFI lets the caller supply liquidity, ADV, beta, or sector weights so the peer pressure index is a portfolio, not a simple average. Rolling OLS beta of own return on that weighted peer mean produces Cont-style cross-asset impact and a residual idiosyncratic return.
Recurrence
Let \(r_t\) be own return, \(f_{t,i}\) peer OFIs, and \(w_{t,i}\) peer weights for \(i=1,\ldots,P\). Let \(W_t = \sum_i w_{t,i}\) over positive finite weights.
Maintain rolling pair statistics over the last \(n\) samples (window) of
\((r_t, \bar f_t)\) and form OLS beta \(\hat\beta_t\). Then
Implementation Notes
The recurrence is implemented in src/rtta/indicator.cpp in
class WeightedMultiPeerOrderFlowImbalance using RollingPairStats. Unlike
equal-weight MultiPeerOrderFlowImbalance, every update requires a weight
vector (or weight matrix in batch).