Skip to main content

​
GLFT Optimal Market Maker

The GLFT strategy is an academic-grade market making engine based on the Gueant-Lehalle-Fernandez-Tapia optimal quoting framework. Instead of relying on heuristic rules, GLFT derives optimal bid/ask placement from a closed-form mathematical solution that minimizes inventory risk while maximizing expected profit.
GLFT is currently in ALPHA. It is a fully functional strategy with advanced features, but parameter defaults may evolve as we gather more production data. Start with a conservative preset and small position sizes.

​
What is GLFT?

The Gueant-Lehalle-Fernandez-Tapia model (published in 2013) solves the optimal market making problem analytically. Given a market maker’s inventory, current volatility, and order arrival rates, it computes the mathematically optimal bid and ask prices that balance two competing goals:
  1. Maximize spread capture - quote tightly to get filled more often
  2. Minimize inventory risk - avoid building a large directional position that could lose money
Unlike the Perp Market Maker which uses configurable heuristic rules (100+ parameters), GLFT uses a small set of inputs to derive spreads from equations. The result is a strategy that adapts automatically to market conditions with fewer knobs to turn.
Academic Foundation: The GLFT model is one of the most widely cited papers in quantitative market making. It provides a rigorous mathematical framework that professional trading firms use as a starting point for their own strategies.

​
How It Works

​
The GLFT Engine

The core engine computes optimal bid/ask offsets from mid-price using four key inputs: 
                    GLFT Optimal Quoting Engine
                    ──────────────────────────
                              │
           ┌──────────────────┼──────────────────┐
           │                  │                  │
      ┌────┴────┐      ┌─────┴─────┐      ┌────┴────┐
      │  Gamma  │      │   Sigma   │      │  Kappa  │
      │  (risk) │      │   (vol)   │      │ (flow)  │
      └────┬────┘      └─────┬─────┘      └────┬────┘
           │                  │                  │
           └──────────────────┼──────────────────┘
                              │
                              ▼
                    ┌─────────────────┐
                    │ Optimal Spread  │
                    │   Calculation   │
                    └────────┬────────┘
                              │
              ┌───────────────┼───────────────┐
              │                               │
        ASK: mid + δa                  BID: mid - δb
        (sell offset)                  (buy offset)
Gamma (risk aversion, default 0.01) - Controls how aggressively the engine penalizes inventory. Higher gamma = wider spreads, less position risk. Sigma (realized volatility) - Measured from recent price data. Higher volatility = wider optimal spreads to compensate for adverse selection risk. Kappa (order arrival rate) - Estimated from orderbook liquidity depth. Denser orderbooks = higher kappa = tighter optimal spreads because fills are more likely.

​
Inventory Skew

The key insight of the GLFT model: quotes should be asymmetric based on inventory. 
NEUTRAL INVENTORY (q = 0):          LONG INVENTORY (q > 0):
Symmetric quotes                    Skewed to reduce position

    ASK: $100.08                        ASK: $100.05  ← tighter (want to sell)
         │                                   │
    MID: $100.00                        MID: $100.00
         │                                   │
    BID: $99.92                         BID: $99.90   ← wider (less buying)


SHORT INVENTORY (q < 0):
Skewed to reduce position

    ASK: $100.10  ← wider (less selling)
         │
    MID: $100.00
         │
    BID: $99.95   ← tighter (want to buy)
This automatic skewing means the strategy naturally works to reduce inventory risk without needing manual intervention.

​
Key Features

​
GLFT Optimal Quoting

Spreads are derived from the closed-form GLFT equations, not from manually tuned heuristics. The engine adapts in real-time as volatility and orderbook conditions change.

​
Inventory Penalty

The gamma parameter controls how strongly the engine penalizes inventory accumulation. At default gamma (0.01), the engine gently skews quotes. Increase gamma to 0.05-0.1 for more aggressive inventory reduction.

​
Regime Detection

The engine detects three market regimes via ADX (Average Directional Index) and adjusts behavior:
RegimeDetectionGamma MultiplierEffect
RangingADX < 200.7xTighter spreads, more quoting
TrendingADX > 25, directional0.7x with trend, 1.5x counterFavors trend direction
VolatileHigh NATR1.3xWider spreads, defensive

​
Recovery Grid

An orderbook-aware DCA system that activates when positions are underwater. Places recovery orders at detected support/resistance levels with martingale sizing. See the Recovery Grid section below for details.

​
Grid Breathing

Dynamic spacing that adapts to drawdown conditions:
  • Drawdown-based: Grid widens as unrealized loss grows (base 150 bps)
  • Per-side breathing: Each side (long/short) breathes independently based on its own P&L
  • Smoothing: Gradual transitions prevent whipsaw (alpha_widen: 0.35, alpha_tighten: 0.1)

​
Mantis Spread

A composite spread signal that combines three microstructure indicators:
  • Orderbook imbalance - Bid vs ask depth ratio
  • Trade flow - Recent buy vs sell aggressor volume
  • Cartea-Jaimungal estimation - Academic arrival rate model

​
Squeeze Detection

Detects short squeeze and long squeeze conditions by monitoring funding rate extremes and open interest changes, then adjusts quoting accordingly.

​
Time Decay

Gradually reduces TP targets on aging positions to prevent indefinite holding:
RegimeMax Loss Time
Ranging320 minutes
Volatile240 minutes
Trending120 minutes
TP decay begins after 15 minutes (900 seconds) and increases with position age. Each side ages independently when salvage_per_side_age is enabled.

​
CJ Estimator

The Cartea-Jaimungal intensity estimator provides a more accurate model of order arrival rates by analyzing recent trade data, improving the kappa parameter estimation beyond simple orderbook depth.

​
Dynamic Sizing

Quote size adjusts based on market conditions:
  • Base size: 10USDTperorder(conservative)/10 USDT per order (conservative) / 25 (aggressive)
  • Max size: 15−15-20 (conservative) / 30−30-40 (aggressive)
  • Scales between base and max based on volatility and fill rates

​
Inventory TP

Take-profit targets tighten exponentially as inventory grows. When your position is small, the bot waits for full TP. As position size increases, it accepts progressively smaller profits to reduce exposure faster.

​
Multi-Symbol Support

Run the GLFT engine across up to 5 symbols simultaneously from a single bot instance. Each symbol maintains independent state, inventory tracking, and recovery grids.

​
Supported Exchanges

BloFin

  • Requires API key, secret, and passphrase
  • Maker fee: 2 bps, Taker fee: 6 bps
  • Supports up to 75x leverage (aggressive preset)
  • WebSocket orders and data streams

Bybit

  • Requires API key and secret only
  • Auto fee detection (no manual fee config needed)
  • 20x leverage (default for all presets)
  • WebSocket order management

​
Available Presets

GLFT ships with 8 presets covering both exchanges in conservative and aggressive configurations, plus multi-symbol variants.
PresetRiskQuote SizeMax PositionLeverageMin Profit
ConservativeLow$10$50020x16 bps
AggressiveHigh$25$2,00075x12 bps
Multi ConservativeLow$10$500/symbol20x16 bps
Multi AggressiveHigh$25$2,000/symbol75x12 bps
All GLFT presets appear in the Suggested Presets section when deploying a bot. Look for the “ALPHA” tag. See the Preset Selection Guide for general preset guidance.

​
Key Parameters

ParameterDescriptionConservativeAggressive
min_profit_bpsMinimum profit target in basis points14-1612
quote_size_usdtBase quote size per order$10$25
max_position_usdtMaximum position per symbol$5001,000−1,000-2,000
leverageLeverage multiplier20x20-75x
gammaRisk aversion (higher = wider spreads)0.010.01
grid.levelsNumber of quote levels per side22
max_drawdown_usdtHard stop loss per symbol100−100-200300−300-500
loop_interval_secondsQuote refresh interval0.5s0.5s

​
Grid Configuration

​
grid.levels
integer
default:"2"
Number of quote levels per side. Level 1 sits closest to mid-price, Level 2 is offset by spacing_bps.
​
grid.spacing_bps
array
default:"[0, 20]"
Basis point offset for each grid level from the GLFT optimal price. Level 1 at 0 bps (optimal), Level 2 at 20 bps wider.
​
grid.size_pct
array
default:"[0.6, 0.4]"
Size allocation per level as fraction of total quote size. 60% at Level 1, 40% at Level 2.
​
grid.regime_adjust
boolean
default:"true"
Enable regime-based grid adjustments. Multiplies grid parameters by regime-specific factors.
​
grid.inv_skew
boolean
default:"true"
Enable inventory-based quote skewing. Reduces quoting on the side that would increase inventory.
​
grid.per_side_breathing
boolean
default:"true"
Each side (long/short) breathes independently based on its own drawdown and duration.

​
Recovery Grid

When a position moves against you, the recovery grid activates to manage the underwater position.

​
How It Works

1

Detection

The bot detects an underwater position and activates recovery mode. Normal grid quoting on the losing side is suppressed (suppress_grid_adds: true).
2

Orderbook Analysis

The orderbook_dca mode scans the orderbook for support (for longs) or resistance (for shorts) walls. It selects the strongest price levels as recovery entry points.
3

Order Placement

Recovery orders are placed at detected levels, with a minimum distance of 1% from the last fill and 1% between each recovery order. Up to 2 recovery levels are placed.
4

Martingale Sizing

Order size scales with unrealized PnL depth to average down more aggressively at larger drawdowns.

​
Martingale Size Scaling

Unrealized PnL %Size MultiplierExample (base $10)
Below 5%Base size$10
5%0.5x position$10 + 50% of current position
10%1.0x position$10 + 100% of current position
20%1.5x position$10 + 150% of current position

​
Safety Features

  • Spike blocking: Pauses DCA during sharp price moves (checks 15m candle data)
  • Dynamic depth: Recovery range calculated from recent 8-hour price action on 15m timeframe
  • Minimum distance: 3% minimum distance from current price to first recovery level
  • Repost threshold: Orders only reposted if price moves more than 0.1% from current placement
The recovery grid is not a guarantee against losses. In strong trends, recovery orders can increase position size into a losing trade. The max_drawdown_usdt hard stop is your ultimate safety net.

​
Risk Warning

Market making involves significant risk. Key risks specific to GLFT:
  • Inventory risk: Despite optimal skewing, strong trends can build large positions before the stop loss triggers
  • Recovery grid risk: Martingale sizing adds to losing positions - in severe moves this increases total loss
  • Leverage risk: Even 20x leverage means a 5% adverse move equals 100% of position value
  • Liquidity risk: On low-liquidity tokens, spreads may not be wide enough to cover adverse selection
  • Alpha status: GLFT is in active development - parameter defaults may change between updates
Only trade with funds you can afford to lose. Start with the conservative preset and monitor closely.

​
Who Is This For?

Academic Traders

Those who want mathematically-derived quoting rather than hand-tuned heuristics

Experienced MMs

Market makers comfortable with inventory risk who want fewer parameters to manage

Multi-Symbol Traders

Those who want to run one bot across multiple symbols simultaneously
Minimum requirements:
  • $500+ account balance (conservative preset)
  • Understanding of market making concepts (spreads, inventory, adverse selection)
  • Willingness to monitor positions, especially during the alpha period
  • BloFin or Bybit exchange account with API access

​
GLFT vs Perp Market Maker

AspectGLFTPerp MM
Quoting methodAcademic equationsHeuristic rules
Parameters~20 key settings100+ parameters
Spread derivationAutomatic from volatility/flowManual base_spread_bps
Inventory managementBuilt-in penalty functionConfigurable skew factor
Recovery systemOrderbook DCA with martingaleXGrid counter scalp
Best forFewer knobs, academic approachFull customization
ComplexityMediumHigh

​
Next Steps

Vaults

Deploy GLFT through managed vaults for hands-off trading

Preset Guide

Compare all presets across strategies and exchanges

Perp Market Maker

Compare with the heuristic-based Perp MM strategy

Bot Management

Learn how to deploy, monitor, and manage your bots