This package implements some common routines for state-space models. Provided algorithms include:
- Kalman filter (
kalman_filter
) - Chandrasekhar recursions (
chand_recursion
): "Using the "Chandrasekhar Recursions" for Likelihood Evaluation of DSGE Models" (2012) - Tempered particle filter (
tempered_particle_filter
): "Tempered Particle Filtering" (2016) - Kalman smoothers:
hamilton_smoother
: James Hamilton, Time Series Analysis (1994)koopman_smoother
: S.J. Koopman, "Disturbance Smoother for State Space Models" (Biometrika, 1993)
- Simulation smoothers:
carter_kohn_smoother
: C.K. Carter and R. Kohn, "On Gibbs Sampling for State Space Models" (Biometrika, 1994)durbin_koopman_smoother
: J. Durbin and S.J. Koopman, "A Simple and Efficient Simulation Smoother for State Space Time Series Analysis" (Biometrika, 2002)
StateSpaceRoutines.jl
is a registered Julia package in the General
registry. To install, open your Julia REPL, type ]
(enter package manager), and run
pkg> add StateSpaceRoutines
StateSpaceRoutines.jl
is currently compatible with Julia 1.x
.
To use StateSpaceRoutines.jl
with Julia v0.7
, please check out tag 0.2.0
. To do this, click on the drop-down menu that reads branch: main
on the left-hand side of the page. Select tags
, then v0.2.0
. If you've already cloned the repo, you can simply run git checkout v0.2.0
.
The StateSpaceRoutines.jl
package is not precompiled by default because when running code in parallel, we want to re-compile
the copy of StateSpaceRoutines.jl
on each processor to guarantee the right version of the code is being used. If users do not
anticipate using parallelism, then users ought to change the first line of src/StateSpaceRoutines.jl
from
isdefined(Base, :__precompile__) && __precompile__(false)
to
isdefined(Base, :__precompile__) && __precompile__(true)
s_{t+1} = C + T*s_t + R*ϵ_t (transition equation)
y_t = D + Z*s_t + u_t (measurement equation)
ϵ_t ∼ N(0, Q)
u_t ∼ N(0, E)
Cov(ϵ_t, u_t) = 0
kalman_filter(y, T, R, C, Q, Z, D, E, s_0 = Vector(), P_0 = Matrix(); outputs = [:loglh, :pred, :filt], Nt0 = 0)
chand_recursion(y, T, R, C, Q, Z, D, E, s_pred = Vector(), P_pred = Matrix(); allout = false, Nt0 = 0, tol = 0.0)
tempered_particle_filter(y, Φ, Ψ, F_ϵ, F_u, s_init; verbose = :high, include_presample = true, fixed_sched = [], r_star = 2, c = 0.3, accept_rate = 0.4, target = 0.4, xtol = 0, resampling_method = :systematic, N_MH = 1, n_particles = 1000, Nt0 = 0, adaptive = true, allout = true, parallel = false)
hamilton_smoother(y, T, R, C, Q, Z, D, E, s_0, P_0; Nt0 = 0)
koopman_smoother(y, T, R, C, Q, Z, D, s_0, P_0, s_pred, P_pred; Nt0 = 0)
carter_kohn_smoother(y, T, R, C, Q, Z, D, E, s_0, P_0; Nt0 = 0, draw_states = true)
durbin_koopman_smoother(y, T, R, C, Q, Z, D, E, s_0, P_0; Nt0 = 0, draw_states = true)
For more information, see the docstring for each function (e.g. enter ?kalman_filter
in the REPL).
All of the provided algorithms can handle time-varying state-space systems. To do this, define regime_indices
, a Vector{Range{Int64}}
of length n_regimes
, where regime_indices[i]
indicates the range of periods t
in regime i
. Let T_i
, R_i
, etc. denote the state-space matrices in regime i
. Then the state space is given by:
s_{t+1} = C_i + T_i*s_t + R_i*ϵ_t (transition equation)
y_t = D_i + Z_i*s_t + u_t (measurement equation)
ϵ_t ∼ N(0, Q_i)
u_t ∼ N(0, E_i)
Letting Ts = [T_1, ..., T_{n_regimes}]
, etc., we can then call the time-varying methods of the algorithms:
kalman_filter(regime_indices, y, Ts, Rs, Cs, Qs, Zs, Ds, Es, s_0 = Vector(), P_0 = Matrix(); outputs = [:loglh, :pred, :filt], Nt0 = 0)
hamilton_smoother(regime_indices, y, Ts, Rs, Cs, Qs, Zs, Ds, Es, s_0, P_0; Nt0 = 0)
koopman_smoother(regime_indices, y, Ts, Rs, Cs, Qs, Zs, Ds, s_0, P_0, s_pred, P_pred; Nt0 = 0)
carter_kohn_smoother(regime_indices, y, Ts, Rs, Cs, Qs, Zs, Ds, Es, s_0, P_0; Nt0 = 0, draw_states = true)
durbin_koopman_smoother(regime_indices, y, Ts, Rs, Cs, Qs, Zs, Ds, Es, s_0, P_0; Nt0 = 0, draw_states = true)
The tempered particle filter is a particle filtering method which can approximate the log-likelihood value implied by a general (potentially non-linear) state space system with the following representation:
s_{t+1} = Φ(s_t, ϵ_t) (transition equation)
y_t = Ψ(s_t) + u_t (measurement equation)
ϵ_t ∼ F_ϵ(∙; θ)
u_t ∼ N(0, E)
Cov(ϵ_t, u_t) = 0
- The documentation and code are located in src/filters/tempered_particle_filter.
- The example is located in docs/examples/tempered_particle_filter
- The paper proposing and analyzing the method is Herbst and Schorfheide (2019)
- For nonlinear smoothing, see LowLevelParticleFilters.jl
Copyright Federal Reserve Bank of New York. You may reproduce, use, modify, make derivative works of, and distribute and this code in whole or in part so long as you keep this notice in the documentation associated with any distributed works. Neither the name of the Federal Reserve Bank of New York (FRBNY) nor the names of any of the authors may be used to endorse or promote works derived from this code without prior written permission. Portions of the code attributed to third parties are subject to applicable third party licenses and rights. By your use of this code you accept this license and any applicable third party license.
THIS CODE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT ANY WARRANTIES OR CONDITIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY WARRANTIES OR CONDITIONS OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, EXCEPT TO THE EXTENT THAT THESE DISCLAIMERS ARE HELD TO BE LEGALLY INVALID. FRBNY IS NOT, UNDER ANY CIRCUMSTANCES, LIABLE TO YOU FOR DAMAGES OF ANY KIND ARISING OUT OF OR IN CONNECTION WITH USE OF OR INABILITY TO USE THE CODE, INCLUDING, BUT NOT LIMITED TO DIRECT, INDIRECT, INCIDENTAL, CONSEQUENTIAL, PUNITIVE, SPECIAL OR EXEMPLARY DAMAGES, WHETHER BASED ON BREACH OF CONTRACT, BREACH OF WARRANTY, TORT OR OTHER LEGAL OR EQUITABLE THEORY, EVEN IF FRBNY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES OR LOSS AND REGARDLESS OF WHETHER SUCH DAMAGES OR LOSS IS FORESEEABLE.