What a systematic volatility trading system looks like end to end:
data → forecasting → strategy → portfolio risk → execution → reconciliation →
monitoring → research feedback.
This is a target, not a description of the box. Most of the
right-hand side does not exist. There is no order manager, no broker adapter,
no ledger, no reconciliation, and no portfolio risk engine — see
Not built. What runs today is the left and middle:
ingest, the warehouse, and the pricing / surface analytics described in
Ingest and pricing.
The control loop. Strategy generation, portfolio risk,
execution, ledger, reconciliation, and active monitoring — none of which
exist today.
Pricing separated from forecasting. Surface fitting and
greeks are current-market analytics. They are not a realized-vol forecast,
and conflating them is the easiest way to think a model has edge.
Delayed research feeds separated from live execution feeds.
The delayed feed is enough for research and sanity checks, and not enough
for order placement or intraday risk.
State made explicit. Internal positions, orders, fills,
and decision logs are first-class components rather than something recovered
from the broker on demand.
A closed feedback loop. Backtests, replay, attribution,
dashboards, and governance feed back into production changes.