Simplex routines, def_sim macro, discrete_var quantile and quadrature methods

New features

• Addition of Quantile and Quadrature methods for discrete_var. Thanks to @Shunsuke-Hori (ref #198 for quadrature and #194 for quantile)
• Add @def_sim macro to simplify the creation of structs to hold simulations and observations from models. See the docstring for more info
• Added routines for working with simplex. Thanks @cc7768 (ref #183 and #178 and #193)
• New functions next_k_array! and k_array_rank. Thanks @Shunsuke-Hori (ref #199)

Improvements/Bug fixes

• In compute_fixed_point, don't warn when convergence is achieved in exactly max_iter iterations. Thanks @a-parida12 (ref #195)
• Use StatsBase.ecdf instead of our own implementation. Thanks @a-parida12 (ref #196)
• type restrictions on 1d arrays in DiscreteDP routines loosened from Vector to AbstractVector (ref f53bfd7)
• Remove 0.0 as candidate solution in solve_discrete_riccati. Thanks @oyamad (ref #185)
• Used femtocleaner to upgrade to Julia v0.6+ syntax

Utility functions, ddp perf, bug fixes

New features

• There are now a handful of common utility functions and their derivatives implemented as types. Thanks @cc7768 in #179

Improvements

• Allow Q matrix to be any subtype of AbstractMatrix (including sparse) when using SA formulation of DDP #181

Bug fixes

• update discrete_var methods to use new conventions in Optim.jl. Thanks @Shunsuke-Hori #182
• Bug in LinInterp #177 when evaluating non-consecutive columns

Discrete VAR MC approximation of Farmer and Toda

New Features

• Added routine to compute a finite-state Markov chain approximation to a VAR(1). Thanks to @Shunsuke-Hori . Algorithm adapted from

Farmer, L. E., & Toda, A. A. (2017).
"Discretizing nonlinear, non‐Gaussian Markov processes with exact conditional moments,"
Quantitative Economics, 8(2), 651-683.

Docs + noisy LSS

New features:

• We can now have noise in the observation equation in the LSS type (see #167). Thanks @vgregory757
• Docstrings have been cleaned up and now render math in generated html (see #168) Thanks @natashawatkins

Internal changes:

• Maximum number of newton iterations in qnwgamma increased from 10 to 25. Fixes an issue @sglyon had in some research code.

Discrete MC estimation

New features:

• Estimation of discrete Markov Chain from observed data (cc @cc7768, ref: #161)
• Method rand!(output::AbstractArray{<:Integer}, ::DiscreteRV) to fill pre-populated array with samples from a DiscreteRV

Changes:

• deprecate draw(::DiscreteRV, ...)in favor of rand(::DiscreteRV, ...)

Code style and matrix LinInterp

Fixes code styling and extends LinInterp to handle multiple functions at one time

Bugfix in MVNSampler

A bugfix release

Improvements

• This fixes a bug in MVNSampler when using covariance matrices will eltype <: Integer or <: Rational{<:Integer}

out with 0.4, in with 0.6

Julia 0.4 has officially been dropped. The 0.9.x release line of this package is the final version supporting Julia 0.4.

New features

• New MVNSampler type that allows you to sample from a multivariate normal distribution with a positive semi-definite covariance matrix. Previously we used the MvNormal type from Distributions.jl, which only supported positive definite matrices. ref #157 thanks @Shunsuke-Hori and @oyamad
• New type LSSMoments that implements the Julia iterator interface (start, next, and done) to replace the former task based routines. See commit: fd303c4

Breaking

• Task based API for the moment sequence of an instance of LSS has been replaced by the LSSMoments type (See above)
• ecdf(x::ECDF, y) is now spelled x(y)

Improvements

• Julia 0.6 is fully supported with no deprecation warnings
• bug fix in estspec.jl. See discussion here
• Type stability improvements to quadrature routines
• type stability in vararg version of gridmake

update

v0.2.0

QuantEcon v0.2.0 [af4a7f85e6]

v0.1.0

Changes:

• Added gth_solve for solving for the stationary distribution of an irreducible Markov transition matrix (stochastic matrix) or transition matrix (generator matrix) A using the Grassmann-Taksar-Heyman (GTH) algorithm
• Renamed DMarkov to MarkovChain. This comes with many fixes and improvements.
• Using Compat.jl to make compatible with v0.3.x and v0.4.x of Julia
• Using internal quadrature methods instead of quadgk from base -- leads to over 125x speedup for solving some models