Class AvellanedaStoikov

java.lang.Object
com.quantfinlib.trading.AvellanedaStoikov
public final class AvellanedaStoikov extends Object
Avellaneda-Stoikov (2008) optimal market-making quotes — the principled version of the inventory skew HftQuoter applies heuristically. Two closed-form answers from one utility-maximization problem:
  1. Where is MY mid? The reservation price shades the market mid against your inventory: r = mid − q·γ·σ²·τ. Long inventory (q > 0) pushes both quotes down — you want to sell, so you make selling attractive and buying not. The shade grows with risk aversion (γ), variance (σ²) and remaining horizon (τ): a big position in a wild market you must carry for hours is worth shading hard;
  2. How wide? The optimal total spread δ = γ·σ²·τ + (2/γ)·ln(1 + γ/κ) balances the volatility cost of holding inventory against the fill-rate cost of quoting wide, where κ is the order-arrival decay (how fast fill intensity drops as you quote away from the touch — bigger κ = thicker flow near the mid = quote tighter). As γ → 0 the spread collapses to the pure liquidity floor 2/κ.

Units contract (same discipline as execution.BenchmarkExecutor.MarketState):

  • priceVariancePerSecond — variance of the PRICE per second, not the return: from the signal engine, Math.pow(mid * sig.volPerSqrtSecond(sym), 2). NaN or negative reads as 0 (the inventory term disables; the liquidity floor keeps quoting sane);
  • horizonSeconds — time to the moment inventory must be flat (the close, the fixing). Continuous 24/5 FX has no terminal time: use a fixed risk horizon (how long you are willing to sit on a position) — the standard practitioner reading of τ;
  • inventory — signed position in UNITS OF THE INSTRUMENT the mid prices (shares of the stock, base-currency units of the pair). This is dimensional, not stylistic: the shade q·γ·σ²·τ is in price units only when q counts the same thing σ² is quoted per — pass round lots instead of shares and the shade silently shrinks 100× while the spread term (which never sees q) is unchanged, gutting the skew that is the model's point. Calibrate γ with the inventory unit fixed first.

Wiring into the fast lane: compute reservationPrice − mid as the inventory skew and optimalHalfSpread(double, double) as the half-spread, and feed both to HftQuoter's skew/spread inputs — the model decides, the quoter executes. Pure static-shape math on primitives: zero allocation, safe on any thread. Like every model here, γ and κ deserve calibration against your own fill data before the output is trusted with size.

  • Constructor Summary

    Constructors
    Constructor
    Description
    AvellanedaStoikov(double gamma, double kappa)
     

  • Method Summary

    Modifier and Type
    Method
    Description
    double
    askQuote(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
    The ask to quote: reservation price plus the optimal half-spread.
    double
    bidQuote(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
    The bid to quote: reservation price minus the optimal half-spread.
    double
    gamma()
     
    double
    kappa()
     
    double
    optimalHalfSpread(double priceVariancePerSecond, double horizonSeconds)
    Half of the optimal total spread: (γ·σ²·τ)/2 + (1/γ)·ln(1 + γ/κ).
    double
    reservationPrice(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
    The inventory-shaded fair value: mid − inventory·γ·σ²·τ.

    Methods inherited from class java.lang.Object

    clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

  • Constructor Details

    • AvellanedaStoikov

      public AvellanedaStoikov(double gamma, double kappa)
      Parameters:
      gamma - risk aversion, e.g. 0.1 (bigger = shade and widen more)
      kappa - fill-intensity decay per unit distance from the touch, e.g. 1.5 (bigger = flow concentrates at the mid)

  • Method Details

    • reservationPrice

      public double reservationPrice(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
      The inventory-shaded fair value: mid − inventory·γ·σ²·τ. Flat inventory or a dead/garbage variance returns the mid itself.

    • optimalHalfSpread

      public double optimalHalfSpread(double priceVariancePerSecond, double horizonSeconds)
      Half of the optimal total spread: (γ·σ²·τ)/2 + (1/γ)·ln(1 + γ/κ). Never below the liquidity floor — even a becalmed market pays for immediacy.

    • bidQuote

      public double bidQuote(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
      The bid to quote: reservation price minus the optimal half-spread.

    • askQuote

      public double askQuote(double mid, double inventory, double priceVariancePerSecond, double horizonSeconds)
      The ask to quote: reservation price plus the optimal half-spread.

    • gamma

      public double gamma()

    • kappa

      public double kappa()