A quantitative trading system for digital asset markets. Strategies validated across three regimes, deployed only after walk-forward confirmation, executed in under twenty milliseconds.
Takumi — the mastery that comes from decades of deliberate practice. The craftsman who refines a single discipline until it becomes second nature. Here, that discipline is translated into hard rules — gates that no strategy passes without earning it.
一 つ の 道 を 究 め るThese are not aspirations — they are the gates every strategy must pass through before it touches capital.
Bull, bear, and chop. Backtest results on training data are disregarded. Only walk-forward results count toward deployment.
Capped at strategy level, aggregated at portfolio level. Correlation-adjusted during high-volatility regimes. No exceptions.
Trailing stops in favor of the position are allowed. Widening stops after entry is the fastest route to structural drawdown.
Automated halt of all strategies. Manual review before redeployment. Protects against regime shifts the models have not yet seen.
Full audit trail from signal to fill. Strategy configs versioned in git. No manual overrides that bypass the record.
Live conditions, zero risk. Validates what backtests miss — slippage, partial fills, queue position, API quirks.
Every strategy has a falsifiable hypothesis and a single mechanical entry rule. No discretion. No override. The trigger fires or it does not.
Sustained directional moves on BTC and ETH perpetuals are preceded by volatility expansion above the 20-day ATR baseline, confirmed by funding-rate neutrality. The most battle-tested strategy in the portfolio.
Short-term price extremes on liquid perps revert when order-book imbalance contradicts the move. Requiring book disagreement with price direction avoids catching falling knives.
Extreme funding rates on perpetual contracts are captured market-neutrally with a delta-hedged spot position, isolating the funding payment as the primary return source.
Temporary price dislocations between spot venues during volatility spikes mean-revert within seconds. Execution is highly sensitive to latency, withdrawal times, and inventory balance.
Cascading liquidations exhaust when large open-interest clusters are cleared. The subsequent bounce is tradeable if cascade termination can be identified in real time.
A modular pipeline engineered from first principles. Market data flows from exchanges into a normalized store; strategies subscribe to signals, a shared risk layer sizes positions, and execution routes orders to venues.
WebSocket streams
Tick-level history
ClickHouse
NATS
Strategy modules
TypeScript + Rust
Position sizing
Kill switches
Smart routing
CEX APIs
Figures pulled directly from the trading database. Unlevered. Net of fees and slippage. Updated every thirty minutes from exchange APIs.
This page documents an in-house quantitative trading system operated as a private company. It is not a solicitation, investment offering, or performance advertisement. Past performance does not predict future results. Quantitative trading involves substantial risk of loss.
A running record of strategy deployments, risk framework changes, and infrastructure work.
Exploring queue dynamics and trade flow at 1–30 second horizons. Tick-level data collection active across Binance and Bybit. Early results show predictive power, but margins are thin after fees and slippage.
Deployed a rolling correlation measure across active pairs. When cross-asset correlation exceeds 0.75, per-strategy size scales down by 30%, reducing drawdown during synchronized digital asset moves.
Hypothesis formalized. Backtesting across two years of liquidation data from Coinglass and exchange APIs. Out-of-sample validation scheduled for May 2026.
The previous Postgres setup became a bottleneck at ~30M rows/day. ClickHouse reduced query latency on common aggregations by roughly 40× and simplified columnar research workflows.
Two months of backtest results passed review. Now running live with simulated fills to verify execution assumptions, particularly around withdrawal times and inventory constraints.