finmath lib Release Notes
Release Notes
3.1.0 (30.09.2017)
MonteCarlo models
 RandomVariableInterface gets an additional method doubleValue() which will return the value of random variable if it is deterministic. So instead of getAverage() you may call average().doubleValue(). This addition is to allow the consistent use of deterministic random variables in analytic model. The advantage of this approach is the possibility of dependency injection (using stochastic automatic differentiation) and the possiblity to use stochastic quantities in analytic models (“stochastic curves”).
Analytic models  Curves
 Introduction of “stochastic curves”. The package net.finmath.analytic is a port of net.finmath.marketdata where all curve object operate on RandomVariableInterface. This allows AAD calibration and “stochastic curves”.
Optimizer
 Introduction of the stochastic Levenberg Marquardt algorithm.
3.0.14 (20.08.2017)
MonteCarlo models
 AbstractModelInterface requires the inverse of the statespacetransform. This enables ProcessEulerScheme to construct a sequential dependency structure (this is useful for AAD).
 ProcessEulerScheme offers the scheme EULER_FUNCTiONAL (constructs a sequential dependency structure (this is useful for AAD))
Valuation using Fourier transforms
– Added discount rate as optional parameter to Heston and Bates model.
3.0.12 (19.08.2017)
Valuation using Fourier transforms
 Added characteristic function of Heston model  allows semianalytic valuation of option prices under Heston model.
 Added characteristic function of Bates model  allows semianalytic valuation of option prices under Bates model.
Support for finmathlib automatic differentiation extensions: Interface chances to RandomVariableInterface and MonteCarlo models
 Minor changes to RandomVariableInterface to support the finmathlib automatic differentiation extensions.
 Minor changes to RandomVariableFactories to support the finmathlib automatic differentiation extensions.
 MonteCarlo models may now take their own factory for RandomVariableInterface objects. This allows injection of AAD capable random varaibles.
 Minor changes to BermdanOption valuation to support the finmathlib automatic differentiation extensions.
Improvements
 Fixed a scaling issue in the conditional expectation estimations, improving the result of products using the Option component.
Other
 Java 8 version is the default in the Eclipse .classpath file.
3.0.0 (27.05.2017)
Java 6 version of finmathlib switched from jodatime to threetenbackport
 The Java 6 version of finmathlib switches from jadotime to threetenbackport for the implementaton of LocalDate. This reduces the difference between the Java 6 code base and the Java 8 code base.
 The Java 8 version is not affected.
 The 2.4.x version using the jodatime version of LocalDate is maintained on the branch named “2.4.x”
Introducing a dedicated class for the conversion between LocalDate and Double.
 Finmath lib allows to use Double to specify dates/times and date/timeintervals. The motivation of this is to allow a simplified “textbooklike” definition of models, which is often sufficient for research purposes.. With respect to industry/practical applications it is often necessary to consider exact datebased measurment of time, e.g. using daterolling and daycount conventions. To achive a clear 1:1 mapping, the library fixes a conversion given by a) a reference date and b) a daycount convention (ACT/365).
 The conversion from LocalDate to Double is centralized in the class FloatingpointDate.
 The motivation for ACT/365 is performance (easy conversion). Apart from this there is no specific reason except the need to make some choice.
2.4.4 (20.05.2017)
Introducing a dedicated class for the conversion between LocalDate and Double.
 Finmath lib allows to use Double to specify dates/times and date/timeintervals. The motivation of this is to allow a simplified “textbooklike” definition of models, which is often sufficient for research purposes.. With respect to industry/practical applications it is often necessary to consider exact datebased measurment of time, e.g. using daterolling and daycount conventions. To achive a clear 1:1 mapping, the library fixes a conversion given by a) a reference date and b) a daycount convention (ACT/365).
 The conversion from LocalDate to Double is centralized in the class FloatingpointDate.
 The motivation for ACT/365 is performance (easy conversion). Apart from this there is no specific reason except the need to make some choice.
2.4.2 (05.02.2017)
Bug fixes and improvements
 Invalidating numeraire caches if model is used with a different (new) process.
 Fixed a failure of date lookup in CapletVolatilityParametric (rare).
2.4.1
MonteCarlo Interest Rate Model: Term Structure Model with Time Dependent Tenor Discretization
 Added sample implementation of a term structure model with time dependent tenor discretization (see http://ssrn.com/abstract=2884699 ).
 Added calibration of term structure model
MonteCarlo Interest Rate Model: LIBOR Market Model
 Refactoring: The objects implementing AbstractLIBORCovarianceModelParametric have become immutable. The method setParameter has been removed and replaced by a method getCloneWithModifiedParameter. This allows a performance increase, since in a calibration it is now safe to reuse parts of the model, given that parts of a parameter (e.g. correlation) has not changed.
 Refactoring: The method getLIBOR(double, double, double) performing the interpoation on the tenor structure has been moved from LIBORModelMonteCarloSimualtionInterface to LIBORModelInterface. This allows to have different models implement different interpolation methods (e.g. LMM versus HullWhite). This chance is also motivated by the introduction of LIBOR models with time dependent tenor discretizations.
Optimizer
 LevenbergMarquardt algorithm will stop if improvement is smaller than given accurarcy. Previously the solver was running more interations than required.
2.4.0
Market Data / Schedule Generation
 Added (optional) endofmonth schedule generation to the ScheduleGenerator.
Note: This includes a small change in the schedule generation when schedules are generated as rolling up/down from a 29th, 30th or 31st.
2.3.6
SABR Model
 Added analytic formulas for SABR skew and SABR curvature.
2.3.4
Market Data / Curve Calibration
 Added valuation of Resettable Cross Currency Swap (aka MarktoMarket Cross Currency Swap).
2.3.0
Dependencies
 Adding jblas 1.2.4 for LinearAlgebra.
Since the implementation of commonsmath matrix solver is sometimes noticable slower, we switched to jblas. Note that jblas is used by default, but you may use commonsmath instead by setting the property net.finmath.functions.LinearAlgebra.isUseApacheCommonsMath to true.
Note that jblas is currently not an OSGi bundle. I try to work on that.
2.2.5
Random Number Generators
 Addes Poisson distribution for jumpdiffusion processes.
 Added class IndependentIncrements allowing to create mixed process increments (Brownian increments and jump process increments). See Merton model for a demo.
Market Data / Interest Rates Curve / MultiCurve
 Added Deposit (may be used as calibration product in multicurve calibration)
 Added FRA (may be used as calibration product in multicurve calibration)
Equity / Single Asset Models
 Added Merton model (MonteCarlo simulation)
 Added Heston model (MonteCarlo simulation)
 Added hedge simulation based on meanvariance hedging (using American MonteCarlo / regression).
Other
 Some demo spreadsheets have been added at finmathspreadsheets.
2.2.0
Dependencies
 Replaced colt1.2.0 by apache commonsmath3.6.1.
Since the implementation of MersenneTwister in commonsmath differs from that in colt, this change will lead to small changes in MonteCarlo valuations. Since the implementation of linear equation solver in commonsmath differes from that in colt, this change will lead to (very) small changes in calibration parameters.
The switch from colt1.2.0 to commonsmath was necessary to obtain OSGi compliant setup.
Analytic Formulas
 Added analytic conversion from lognormal to normal (ATM) volatility.
2.1.1
MonteCarlo Simulation of Interest Rate Models
HullWhite Model
 Improved implementation and unit testing. Implementation now uses Browninan motion as a factory for RandomVariableInterface objects.
2.1.0
MonteCarlo Simulation of Interest Rate Models
Interfaces
 From the interface LIBORMarketModelInterface the narrower interface LIBORModelInterface has been extracted, leaving the methods related to the covariance model to LIBORMarketModelInterface only. The method getModel() of LIBORModelMonteCarloSimulationInterface now returns a LIBORModelInterface interface only. In case your code requires access to the covariance model, you have to check for LIBORMarketModelInterface (see the code of getValue in SwaptionAnalyticApproximation for an example).
HullWhite Model
 Added implementation of a MonteCarlo simulation of the HullWhite model.
LIBOR Market Model
 Added LIBOR Market Model (LMM) local volatiltiy model to generate HullWhite model dynamic in an LMM
2.0.3
 Some internals may be configured via Java system properties.
 A bug introduce in 2.0.0 in DayCountConvention_30U_360 in the Java 6 branch has been fixed
2.0.2
 The CalibrationItem of CalibratedCurves may now carry a symbol to create a shifted model (for calculation of sensitivities using finite differences).
2.0.0

API Change: The type java.util.Calendar has been replaced by LocalDate:
 For Java 6 sources (src/main/java6): Replaced Calendar and Date by org.joda.time.LocalDate.
 For Java 8 sources (src/main/java): Replaced Calendar and Date by java.time.LocalDate, contributed by William Wong.

API Change: The DateIndex in net.finmath.montecarlo.interestrate.products.indices now returns the month according to java.time.Month, i.e., January = 1, February = 2, etc.
1.3.6
Swaps
 Additional constructor for Swap (using SwapLegs).
Monte Carlo
 BrownianMotion allows to use a custom AbstractRandomVariableFactory. Useful to switch to single precision floating point numbers (to save memory).
 ProcessEulerScheme has an addition constructor (to directly construct a predictor corrector scheme).
 Swaption is now compatible multicurve LMM (using collateral curve).
Exposure
 Exposure has been renamed to ExposureEstimator (since it is just an estimator, see the corresponding unit test for an application).
 Improved the unit test ExposureTest.
Volatility
 Added SABR analytic approximations to net.finmath.functions.AnalyticFormulas
 BrownianMotion allows to use a custom AbstractRandomVariableFactory. Useful to switch to single precision floating point numbers (to save memory).
Other
 Constructor for trapezoidal rule using equidistant grid.
1.3.5
 Added trapezoidal rule to net.finmath.integration and corresponding unit test.
1.3.4 and earlier
Please check the subversion or git log messages.