HODL LIMITED treats trading as a constrained guidance problem. Our AI continuously computes the optimal target state, then guides the market from where it is — the observed state — through legal intermediate states, monotonically, to that target. Market making and multi-strategy arbitrage, in the lineage of engineering cybernetics.
A ballistic missile never flies straight at its target — it passes through a series of controlled intermediate states and converges with precision. Trading is the same problem. We don't react tick by tick; we compute a target state and guide the order book to it, every cycle, monotonically upward.
On-chain flows, cross-venue order books and real-time sentiment fuse into one signal layer. The model outputs, in real time, the single optimal target state — KPI floor + risk + alpha — for this instant.
When one step can't reach the target, we plan a string of legal intermediate states — constrained state search with receding-horizon replanning (MPC). Every step is legal, and every step converges upward toward the target.
Speed is the moat. Our purpose-built small model decides in milliseconds — fast enough to re-observe, re-plan and re-converge every cycle, before the market drifts away from the target.
Execute → reconcile the observed state → recompute the target → re-plan. Every cycle feeds the next. The system never sits still; it iterates toward the target — which is exactly where AI lives.
Named after Tsien Hsue-shen (钱学森), the father of engineering cybernetics. Purpose-built and deliberately small — 50–1000× smaller than general-purpose LLMs. Small is fast; fast is the edge.
We don't ask "what does the market look like now." We ask "what should the target state be," and "how do we converge to it at the lowest cost, monotonically, without ever breaking a constraint."
In 1954 Tsien Hsue-shen (钱学森) wrote Engineering Cybernetics — the science of guiding a system from any state to a target, under hard constraints, with proof of stability. Seventy years on, it is still the most precise language we have for trading. We map it, one-to-one, onto the quant canon — then put AI inside every loop.
We cast market, inventory and risk as one state vector. Kalman / Wiener filtering — the predictive filtering cybernetics formalized — recovers the true latent state from noisy ticks; our small models learn its nonlinear observation map.
Market making is a Hamilton–Jacobi–Bellman problem (Avellaneda–Stoikov): the optimal bid/ask skew balancing inventory risk against spread capture. AI approximates the value function in real time, tick by tick.
Filling size without moving the market is an optimal-control problem (Almgren–Chriss / Pontryagin): minimize implementation shortfall by trading off impact against timing risk — a planned trajectory, not a single order.
Inventory skew is steered by closed-loop feedback; Lyapunov stability keeps exposure bounded and continuously pulled back to target. The book may drift — it cannot run away.
Each cycle: observe → plan a string of legal intermediate states → execute only the first step → re-observe → re-plan. Model Predictive Control is the loop that binds cybernetics to live trading.
Regimes shift; the controller adapts its gains and stays robust to model error and adversarial flow. AI detects the regime — cybernetics guarantees the system degrades gracefully, never catastrophically.
Control theory tells us where to go and proves we will get there. The quant canon gives us the equations. AI solves them, online, in milliseconds. That triple — Tsien × quant × AI — is HODL's edge.
At its core HODL is an asset manager. Market making and arbitrage are the engines; disciplined portfolio construction, capital allocation and risk-adjusted absolute return are the product. Every pillar runs on the same cybernetic guidance loop.
Two-sided quoting guided to a target state across spot and derivatives — tighter spreads, deeper books, healthier markets.
Same asset, price gaps across venues — captured in real time.
Harvesting perpetual funding spreads with hedged exposure.
Volatility surface and put-call-parity dislocations, captured with hedged legs.
Spot-vs-futures basis captured with delta-neutral books.
Mean-reversion and pairs, steered by the guidance loop.
Beyond crypto — equities and stock-futures basis, one unified engine.
The strategies above are assembled into managed portfolios — engineered for risk-adjusted, market-neutral absolute return, not directional bets.
Uncorrelated strategies blended into one book to smooth the equity curve.
Market-neutral by design; we optimize Sharpe and drawdown, not raw P&L.
Capital is steered across strategies, venues and assets by the control loop.
Segregated managed accounts and bespoke mandates for institutions and projects.
Active market-making and arbitrage across centralized venues — in both HK-regulated and global markets.
Founded by traders and engineers, HODL has been trading since 2020 and was incorporated in 2023. We operate as a proprietary market maker and multi-strategy arbitrage desk, combining institutional-grade risk management with AI research at our core.
Started proprietary trading.
HODL LIMITED incorporated.
Scaled to 8+ venues, multi-asset arbitrage, in-house TSIEN engine.
Pre-trade & real-time limits with automated kill-switches.
Inventory & exposure control under a control-theory framework.
Segregated capital and fully audited execution trails.
For exchanges, token issuers and institutions seeking AI-driven, risk-managed liquidity.