Class HullWhiteModelWithDirectSimulation

  • All Implemented Interfaces:
    IndependentModelParameterProvider, LIBORModel, TermStructureModel, ProcessModel

    public class HullWhiteModelWithDirectSimulation
    extends AbstractProcessModel
    implements LIBORModel
    Implements a Hull-White model with time dependent mean reversion speed and time dependent short rate volatility. Note: This implementation is for illustrative purposes. For a numerically equivalent, more efficient implementation see HullWhiteModel. Please use HullWhiteModel for real applications.

    Model Dynamics

    The Hull-While model assumes the following dynamic for the short rate: \[ d r(t) = ( \theta(t) - a(t) r(t) ) d t + \sigma(t) d W(t) \text{,} \quad r(t_{0}) = r_{0} \text{,} \] where the function \( \theta \) determines the calibration to the initial forward curve, \( a \) is the mean reversion and \( \sigma \) is the instantaneous volatility. The dynamic above is under the equivalent martingale measure corresponding to the numeraire \[ N(t) = \exp\left( \int_0^t r(\tau) \mathrm{d}\tau \right) \text{.} \] The main task of this class is to provide the risk-neutral drift and the volatility to the numerical scheme (given the volatility model), simulating \( r(t_{i}) \). The class then also provides and the corresponding numeraire and forward rates (LIBORs).

    Time Discrete Model

    Assuming piecewise constant coefficients (mean reversion speed \( a \) and short rate volatility \( \sigma \) the class specifies the drift and factor loadings as piecewise constant functions for an Euler-scheme. The class provides the exact Euler step for the short rate r. More specifically (assuming a constant mean reversion speed \( a \) for a moment), considering \[ \Delta \bar{r}(t_{i}) = \frac{1}{t_{i+1}-t_{i}} \int_{t_{i}}^{t_{i+1}} d r(t) \] we find from \[ \exp(-a t) \ \left( \mathrm{d} \left( \exp(a t) r(t) \right) \right) \ = \ a r(t) + \mathrm{d} r(t) \ = \ \theta(t) \mathrm{d}t + \sigma(t) \mathrm{d}W(t) \] that \[ \exp(a t_{i+1}) r(t_{i+1}) - \exp(a t_{i}) r(t_{i}) \ = \ \int_{t_{i}}^{t_{i+1}} \left[ \exp(a t) \theta(t) \mathrm{d}t + \exp(a t) \sigma(t) \mathrm{d}W(t) \right] \] that is \[ r(t_{i+1}) - r(t_{i}) \ = \ -(1-\exp(-a (t_{i+1}-t_{i})) r(t_{i}) + \int_{t_{i}}^{t_{i+1}} \left[ \exp(-a (t_{i+1}-t)) \theta(t) \mathrm{d}t + \exp(-a (t_{i+1}-t)) \sigma(t) \mathrm{d}W(t) \right] \] Assuming piecewise constant \( \sigma \) and \( \theta \), being constant over \( (t_{i},t_{i}+\Delta t_{i}) \), we thus find \[ r(t_{i+1}) - r(t_{i}) \ = \ \frac{1-\exp(-a \Delta t_{i})}{a \Delta t_{i}} \left( ( \theta(t_{i}) - a \bar{r}(t_{i})) \Delta t_{i} \right) + \sqrt{\frac{1-\exp(-2 a \Delta t_{i})}{2 a \Delta t_{i}}} \sigma(t_{i}) \Delta W(t_{i}) \] . In other words, the Euler scheme is exact if the mean reversion \( a \) is replaced by the effective mean reversion \( \frac{1-\exp(-a \Delta t_{i})}{a \Delta t_{i}} a \) and the volatility is replaced by the effective volatility \( \sqrt{\frac{1-\exp(-2 a \Delta t_{i})}{2 a \Delta t_{i}}} \sigma(t_{i}) \). In the calculations above the mean reversion speed is treated as a constants, but it is straight forward to see that the same holds for piecewise constant mean reversion speeds, replacing the expression \( a \ t \) by \( \int_{0}^t a(s) \mathrm{d}s \).

    Calibration

    The drift of the short rate is calibrated to the given forward curve using \[ \theta(t) = \frac{\partial}{\partial T} f(0,t) + a(t) f(0,t) + \phi(t) \text{,} \] where the function \( f \) denotes the instantanenous forward rate and \( \phi(t) = \frac{1}{2} a \sigma^{2}(t) B(t)^{2} + \sigma^{2}(t) B(t) \frac{\partial}{\partial t} B(t) \) with \( B(t) = \frac{1-\exp(-a t)}{a} \).

    Volatility Model

    The Hull-White model is essentially equivalent to LIBOR Market Model where the forward rate normal volatility \( \sigma(t,T) \) is given by \[ \sigma(t,T_{i}) \ = \ (1 + L_{i}(t) (T_{i+1}-T_{i})) \sigma(t) \exp(-a (T_{i}-t)) \frac{1-\exp(-a (T_{i+1}-T_{i}))}{a (T_{i+1}-T_{i})} \] (where \( \{ T_{i} \} \) is the forward rates tenor time discretization (note that this is the normal volatility, not the log-normal volatility). Hence, we interpret both, short rate mean reversion speed and short rate volatility as part of the volatility model. The mean reversion speed and the short rate volatility have to be provided to this class via an object implementing ShortRateVolatilityModel.
    Version:
    1.2
    Author:
    Christian Fries
    See Also:
    ShortRateVolatilityModel, HullWhiteModel
    • Constructor Detail

      • HullWhiteModelWithDirectSimulation

        public HullWhiteModelWithDirectSimulation​(TimeDiscretization liborPeriodDiscretization,
                                                  AnalyticModel analyticModel,
                                                  ForwardCurve forwardRateCurve,
                                                  DiscountCurve discountCurve,
                                                  ShortRateVolatilityModel volatilityModel,
                                                  Map<String,​?> properties)
        Creates a Hull-White model which implements LIBORMarketModel.
        Parameters:
        liborPeriodDiscretization - The forward rate discretization to be used in the getLIBOR method.
        analyticModel - The analytic model to be used (currently not used, may be null).
        forwardRateCurve - The forward curve to be used (currently not used, - the model uses disocuntCurve only.
        discountCurve - The disocuntCurve (currently also used to determine the forward curve).
        volatilityModel - The volatility model specifying mean reversion and instantaneous volatility of the short rate.
        properties - A map specifying model properties (currently not used, may be null).
    • Method Detail

      • getNumberOfComponents

        public int getNumberOfComponents()
        Description copied from interface: ProcessModel
        Returns the number of components
        Specified by:
        getNumberOfComponents in interface ProcessModel
        Returns:
        The number of components
      • getNumberOfFactors

        public int getNumberOfFactors()
        Description copied from interface: ProcessModel
        Returns the number of factors m, i.e., the number of independent Brownian drivers.
        Specified by:
        getNumberOfFactors in interface ProcessModel
        Returns:
        The number of factors.
      • applyStateSpaceTransform

        public RandomVariable applyStateSpaceTransform​(MonteCarloProcess process,
                                                       int timeIndex,
                                                       int componentIndex,
                                                       RandomVariable randomVariable)
        Description copied from interface: ProcessModel
        Applies the state space transform fi to the given state random variable such that Yi → fi(Yi) =: Xi.
        Specified by:
        applyStateSpaceTransform in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        timeIndex - The time index (related to the model times discretization).
        componentIndex - The component index i.
        randomVariable - The state random variable Yi.
        Returns:
        New random variable holding the result of the state space transformation.
      • applyStateSpaceTransformInverse

        public RandomVariable applyStateSpaceTransformInverse​(MonteCarloProcess process,
                                                              int timeIndex,
                                                              int componentIndex,
                                                              RandomVariable randomVariable)
        Description copied from interface: ProcessModel
        Applies the inverse state space transform f-1i to the given random variable such that Xi → f-1i(Xi) =: Yi.
        Specified by:
        applyStateSpaceTransformInverse in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        timeIndex - The time index (related to the model times discretization).
        componentIndex - The component index i.
        randomVariable - The state random variable Xi.
        Returns:
        New random variable holding the result of the state space transformation.
      • getInitialState

        public RandomVariable[] getInitialState​(MonteCarloProcess process)
        Description copied from interface: ProcessModel
        Returns the initial value of the state variable of the process Y, not to be confused with the initial value of the model X (which is the state space transform applied to this state value.
        Specified by:
        getInitialState in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        Returns:
        The initial value of the state variable of the process Y(t=0).
      • getNumeraire

        public RandomVariable getNumeraire​(MonteCarloProcess process,
                                           double time)
                                    throws CalculationException
        Description copied from interface: ProcessModel
        Return the numeraire at a given time index. Note: The random variable returned is a defensive copy and may be modified.
        Specified by:
        getNumeraire in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        time - The time t for which the numeraire N(t) should be returned.
        Returns:
        The numeraire at the specified time as RandomVariable
        Throws:
        CalculationException - Thrown if the valuation fails, specific cause may be available via the cause() method.
      • getDrift

        public RandomVariable[] getDrift​(MonteCarloProcess process,
                                         int timeIndex,
                                         RandomVariable[] realizationAtTimeIndex,
                                         RandomVariable[] realizationPredictor)
        Description copied from interface: ProcessModel
        This method has to be implemented to return the drift, i.e. the coefficient vector
        μ = (μ1, ..., μn) such that X = f(Y) and
        dYj = μj dt + λ1,j dW1 + ... + λm,j dWm
        in an m-factor model. Here j denotes index of the component of the resulting process. Since the model is provided only on a time discretization, the method may also (should try to) return the drift as \( \frac{1}{t_{i+1}-t_{i}} \int_{t_{i}}^{t_{i+1}} \mu(\tau) \mathrm{d}\tau \).
        Specified by:
        getDrift in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        timeIndex - The time index (related to the model times discretization).
        realizationAtTimeIndex - The given realization at timeIndex
        realizationPredictor - The given realization at timeIndex+1 or null if no predictor is available.
        Returns:
        The drift or average drift from timeIndex to timeIndex+1, i.e. \( \frac{1}{t_{i+1}-t_{i}} \int_{t_{i}}^{t_{i+1}} \mu(\tau) \mathrm{d}\tau \) (or a suitable approximation).
      • getFactorLoading

        public RandomVariable[] getFactorLoading​(MonteCarloProcess process,
                                                 int timeIndex,
                                                 int componentIndex,
                                                 RandomVariable[] realizationAtTimeIndex)
        Description copied from interface: ProcessModel
        This method has to be implemented to return the factor loadings, i.e. the coefficient vector
        λj = (λ1,j, ..., λm,j) such that X = f(Y) and
        dYj = μj dt + λ1,j dW1 + ... + λm,j dWm
        in an m-factor model. Here j denotes index of the component of the resulting process.
        Specified by:
        getFactorLoading in interface ProcessModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        timeIndex - The time index (related to the model times discretization).
        componentIndex - The index j of the driven component.
        realizationAtTimeIndex - The realization of X at the time corresponding to timeIndex (in order to implement local and stochastic volatlity models).
        Returns:
        The factor loading for given factor and component.
      • getRandomVariableForConstant

        public RandomVariable getRandomVariableForConstant​(double value)
        Description copied from interface: ProcessModel
        Return a random variable initialized with a constant using the models random variable factory.
        Specified by:
        getRandomVariableForConstant in interface ProcessModel
        Parameters:
        value - The constant value.
        Returns:
        A new random variable initialized with a constant value.
      • getForwardRate

        public RandomVariable getForwardRate​(MonteCarloProcess process,
                                             double time,
                                             double periodStart,
                                             double periodEnd)
                                      throws CalculationException
        Description copied from interface: TermStructureModel
        Returns the time \( t \) forward rate on the models forward curve. Note: It is guaranteed that the random variable returned by this method is \( \mathcal{F}_{t} ) \)-measurable.
        Specified by:
        getForwardRate in interface TermStructureModel
        Parameters:
        process - The discretization process generating this model. The process provides call backs for TimeDiscretization and allows calls to getProcessValue for timeIndices less or equal the given one.
        time - The evaluation time.
        periodStart - The period start of the forward rate.
        periodEnd - The period end of the forward rate.
        Returns:
        The forward rate.
        Throws:
        CalculationException - Thrown if model fails to calculate the random variable.
      • getNumberOfLibors

        public int getNumberOfLibors()
        Description copied from interface: LIBORModel
        Get the number of LIBORs in the LIBOR discretization.
        Specified by:
        getNumberOfLibors in interface LIBORModel
        Returns:
        The number of LIBORs in the LIBOR discretization
      • getLiborPeriod

        public double getLiborPeriod​(int timeIndex)
        Description copied from interface: LIBORModel
        The period start corresponding to a given forward rate discretization index.
        Specified by:
        getLiborPeriod in interface LIBORModel
        Parameters:
        timeIndex - The index corresponding to a given time (interpretation is start of period)
        Returns:
        The period start corresponding to a given forward rate discretization index.
      • getLiborPeriodIndex

        public int getLiborPeriodIndex​(double time)
        Description copied from interface: LIBORModel
        Same as java.util.Arrays.binarySearch(liborPeriodDiscretization,time). Will return a negative value if the time is not found, but then -index-1 corresponds to the index of the smallest time greater than the given one.
        Specified by:
        getLiborPeriodIndex in interface LIBORModel
        Parameters:
        time - The period start.
        Returns:
        The index corresponding to a given time (interpretation is start of period)
      • getAnalyticModel

        public AnalyticModel getAnalyticModel()
        Description copied from interface: TermStructureModel
        Return the associated analytic model, a collection of market date object like discount curve, forward curve and volatility surfaces.
        Specified by:
        getAnalyticModel in interface TermStructureModel
        Returns:
        The associated analytic model.
      • getShortRateConditionalVariance

        public double getShortRateConditionalVariance​(double time,
                                                      double maturity)
        Calculates the variance \( \mathop{Var}(r(t) \vert r(s) ) \), that is \( \int_{s}^{t} \sigma^{2}(\tau) \exp(-2 \cdot \int_{\tau}^{t} a(u) \mathrm{d}u ) \ \mathrm{d}\tau \) where \( a \) is the meanReversion and \( \sigma \) is the short rate instantaneous volatility.
        Parameters:
        time - The parameter s in \( \int_{s}^{t} \sigma^{2}(\tau) \exp(-2 \cdot \int_{\tau}^{t} a(u) \mathrm{d}u ) \ \mathrm{d}\tau \)
        maturity - The parameter t in \( \int_{s}^{t} \sigma^{2}(\tau) \exp(-2 \cdot \int_{\tau}^{t} a(u) \mathrm{d}u ) \ \mathrm{d}\tau \)
        Returns:
        The conditional variance of the short rate, \( \mathop{Var}(r(t) \vert r(s) ) \).
      • getIntegratedBondSquaredVolatility

        public double getIntegratedBondSquaredVolatility​(double time,
                                                         double maturity)