Development of AI-based Market Liquidity Analysis Model
Liquidity — ability of market to absorb orders without significant price movement. For a trader, this is transaction costs: how expensive to execute a position of needed size. For risk manager — how quickly to exit position in a crisis. AI-model evaluates liquidity in real-time and forecasts its changes.
Measuring Liquidity
Bid-Ask Spread: Simplest measure. Relative spread = (Ask - Bid) / Mid. For liquid NYSE stocks: 1-5 bp. For less liquid: 50-200+ bp.
Kyle's Lambda (Price Impact):
ΔPrice = λ × OrderFlow
λ = regression coefficient (price change per unit of signed order flow)
High λ → market reacts quickly to orders → low liquidity.
Amihud Illiquidity Ratio:
ILLIQ = (1/T) × Σ |R_t| / Volume_t
Daily return per unit of trading volume. Standard in academic literature.
Effective Spread and Realized Spread:
- Effective: 2 × |Trade Price - Mid Price| — real cost of aggression
- Realized: 5 minutes after trade — how much market maker actually earned
ML Model for Liquidity Forecasting
Target: bid-ask spread in 15/30/60 minutes, or λ (price impact coefficient).
Features:
| Category | Features |
|---|---|
| Current liquidity | Spread, book depth at 5 levels, quote volume |
| Market activity | Trading volume, trade count, intertrade time |
| Volatility | Realized vol 5/15/60 min, ATR |
| Market regime | VIX, CDS spreads, funding rates |
| Time features | Time of day, day of week, pre/post market |
| News / events | Earnings, macro releases (economic calendar) |
Model: LightGBM Regressor. Gradient boosting works well with tabular liquidity features. MAPE 8-15% for 15-minute forecast — achievable result.
Intraday Liquidity Patterns
Liquidity has stable intraday patterns:
U-shaped curve:
- Market open (9:30-10:00 ET): high spread, thin book
- Lunch (12:00-13:30 ET): minimum volume, worst liquidity
- Close (15:30-16:00 ET): maximum volume, best liquidity
This means: large institutional order should execute closer to close, avoiding open auction.
Event-driven liquidity collapse: News, earnings, FOMC announcements — 5-10 minutes before event market makers remove quotes. Spread widens 5-20×. Model should predict these "liquidity windows".
Measuring Market Impact
Linear Impact Model:
Market Impact = κ × (Order_Size / ADV)^α × Volatility × Sign
κ ≈ 0.1-0.3 (depends on market)
α ≈ 0.5 (square root impact — empirical law)
ADV = Average Daily Volume
Almgren-Chriss Execution Model: Optimal trade schedule to minimize expected impact with time constraint:
Optimal_trajectory = f(volatility, market_impact_params, risk_aversion, T)
ML approach: training on historical execution data with real impact. Can predict impact better than analytical models, especially in non-standard market conditions.
Crisis Liquidity Prediction
During market stress, liquidity evaporates nonlinearly. Task: predict probability of liquidity crisis in next N hours.
Indicators of upcoming crisis:
- Sudden widening of cross-asset correlations (correlation spike)
- Simultaneous liquidity deterioration across multiple asset classes
- CDS spread widening in financial sector
- TED spread (LIBOR - T-bill rate)
- Repo market stress (overnight rate spikes)
Model: Random Forest Classifier. Target: liquidity shock (spread > 3σ from 90-day average) within 24 hours. AUC 0.72-0.80 on historical stress events.
Application in Trading
Execution Optimization:
- Real-time: when and how to execute order
- Liquidity score → choose TWAP/VWAP/IS algorithm
- Adaptive execution: slow down when liquidity deteriorates
Risk Management:
- Liquidity-adjusted VaR: accounts for cost of exiting position
- Position limits: limit position size relative to forecasted liquidity
- Exit stress test: how many days to exit without significant impact under normal and stressed liquidity
Portfolio construction: Include liquidity constraints: don't take positions > X% of ADV, diversify by liquidity.
Timeline: basic liquidity-metrics + intraday pattern model — 3-4 weeks. Full system with market impact prediction, liquidity crisis detection and execution optimization — 3-4 months.