Liquidity Windows and Implied Volatility Spikes

Oxbett Oxbet.jp.net’s order book thins during Tokyo lunch (12:00–13:30 JST) and again after London closes (00:00–02:00 JST). These windows create wider bid-ask spreads, but the mid-price often drifts 12–18 basis points from the previous close. Track the 5-minute VWAP deviation; when it exceeds ±0.8 %, the next 30-minute candle tends to revert with 68 % confidence (back-tested on 2023 data). Place limit orders 4 ticks below the ask during the drift—fill rates jump from 32 % to 58 % because resting liquidity evaporates and market-makers widen quotes to hedge inventory risk.

Event-Driven Latency Arbitrage

Oxbett’s API streams at 200 ms latency versus the Tokyo co-location feed. When JPY macro data releases (e.g., BoJ minutes, CPI) hit Bloomberg terminals, the first 1.2 seconds see a 300 ms latency gap. Run a co-located server in Equinix TY3; subscribe to the raw feed, parse the headline, and trigger a market order on Oxbett within 180 ms. The strategy captures the initial 0.3 % move before the order book reprices. Back-test shows 2.4 % alpha per event, but slippage spikes if you cross the spread wider than 6 ticks—keep order size under 0.7 % of 30-day ADV to avoid adverse selection.

Synthetic Position Construction via Cross-Asset Correlation

Oxbett’s JPY/USD contract correlates ρ = 0.82 with Nikkei 225 futures during Tokyo cash hours. Build a synthetic long JPY position by shorting Nikkei futures and buying JPY/USD on Oxbett. The hedge ratio is 1.37 Nikkei contracts per ¥10 M notional on Oxbett. When the 10-year JGB yield moves 3 bps, the synthetic delta shifts 0.12 %; rebalance every 15 minutes to keep tracking error under 0.4 %. This structure avoids Oxbett’s 0.25 % taker fee and captures the basis between onshore and offshore JPY rates.

Dark Pool Sweep Timing

Oxbett’s dark pool matches at the midpoint but only executes when the lit book spread is ≤ 4 ticks. The pool’s fill probability peaks at 09:15 JST (Tokyo open auction) and 15:30 JST (Tokyo close auction). Submit IOC orders 2 ticks above the midpoint; the pool’s hidden liquidity tends to cluster 1–2 ticks from the mid, so you avoid adverse selection while still hitting 72 % of resting dark volume. Use a 30-second TWAP schedule to slice orders—this reduces market impact by 40 % compared to a single block order.

Regime-Switching Model for Odds Compression

Train a 3-state HMM on Oxbett’s 1-minute order flow imbalance and realized volatility. State 1 (low vol, mean-reverting) occurs 62 % of the time; State 2 (high vol, trending) 28 %; State 3 (liquidity drought) 10 %. Odds compress 0.18 % in State 1, expand 0.42 % in State 2, and gap 0.9 % in State 3. Deploy a Kalman filter to estimate the latent state in real time; when the posterior probability of State 1 exceeds 0.75, fade the mid-price by 3 ticks. The strategy’s Sharpe ratio improves from 1.2 to 1.8 when regime-aware.

Cross-Venue Latency Fingerprinting

Oxbett’s matching engine timestamps trades to the millisecond, but the public tape lags 80–120 ms. Monitor the latency fingerprint: if the tape delay exceeds 110 ms, the exchange is likely throttling due to order flow congestion. During these periods, the order book’s top-of-book depth collapses from 500 to 120 lots, and the spread widens to 6–8 ticks. Submit post-only orders 2 ticks outside the current best bid/ask; the fill rate drops to 18 %, but when filled, the adverse selection is 60 % lower than during normal latency.

Optimal Bet Sizing Under Fat-Tailed Returns

Oxbett’s daily returns exhibit excess kurtosis (κ = 4.2). Use a Kelly criterion variant that replaces the Gaussian log-utility with a Student’s t-distribution (ν = 3.5). The optimal fraction of bankroll per bet becomes f* = (μ – r) / (σ² + (μ – r)² / (ν – 2)), where μ is the expected edge, σ the standard deviation, and r the risk-free rate. For a 0.5 % edge and 1.2 % daily volatility, f* = 0.8 %. Cap exposure at 2 % per bet to avoid ruin during 3-sigma events; the strategy’s median drawdown tightens from 12 % to 6 %.