Absent queue model: no LOB depth awareness; fills treated without queue position logic
VPIN not incorporated: no toxicity filter on incoming flow
Results pre-rebate; maker rebate would improve net PnL
Reservation price:
$r = s - q \cdot \gamma \cdot \sigma^2 \cdot T$
Optimal spread:
$\delta = \gamma \sigma^2 T + \dfrac{2}{\gamma} \ln\!\left(1 + \dfrac{\gamma}{\kappa}\right)$
Fill intensity (MLE calibrated):
$\lambda(\delta) = A \cdot e^{-\kappa \delta}$
γ = 0.1 (risk aversion) · T = 1/24 yr (vol-adaptive horizon) · κ calibrated via MLE on real fill tape
Panel B: Live Model Demonstration
Live
⚠ Instrument Note: Research data = BTCUSD perpetual futures (Coincall, Jun 14–Jun 23 2026).
This live demo runs on BTC-USD spot (Coinbase Advanced Trade). Different instruments, different microstructure.
Parameters are recalibrated from the live trade stream. Results are illustrative only.