adding new option 'fast_kalman_filter' implementing Ed Herbst 2012 approach

doc/dynare.texi

@@ -5527,6 +5527,12 @@ from the likelihood computations (for details see @cite{Durbin and Koopman (2012
 singularity is encountered, Dynare by default automatically switches to the univariate Kalman filter for this parameter draw. This behavior can be changed via the
 @ref{use_univariate_filters_if_singularity_is_detected} option.
 
+@item fast_kalman_filter
+@anchor{fast_kalman_filter} Select the fast Kalman filter using Chandrasekhar
+recursions as described by @cite{Herbst, 2012}. This setting is only used with
+@code{kalman_algo=1} or @code{kalman_algo=3}. It is not yet compatible with
+@code{analytical_derivation}.
+
 @item kalman_tol = @var{DOUBLE}
 @anchor{kalman_tol} Numerical tolerance for determining the singularity of the covariance matrix of the prediction errors during the Kalman filter (minimum allowed reciprocal of the matrix condition number). Default value is @code{1e-10}
 
@@ -13099,6 +13105,11 @@ Hansen, Nikolaus and Stefan Kern (2004): ``Evaluating the CMA Evolution Strategy
 on Multimodal Test Functions''. In: @i{Eighth International Conference on Parallel 
 Problem Solving from Nature PPSN VIII, Proceedings}, Berlin: Springer, 282--291 
 
+@item
+Edward Herbst (2015):
+``Using the “Chandrasekhar Recursions” for Likelihood Evaluation of DSGE
+Models,'' @i{Computational Economics}, 45(4), 693--705.
+
 @item
 Ireland, Peter (2004): ``A Method for Taking Models to the Data,''
 @i{Journal of Economic Dynamics and Control}, 28, 1205--26

matlab/dsge_likelihood.m

@@ -659,6 +659,13 @@ if ((kalman_algo==1) || (kalman_algo==3))% Multivariate Kalman Filter
         if DynareOptions.block
             [err, LIK] = block_kalman_filter(T,R,Q,H,Pstar,Y,start,Z,kalman_tol,riccati_tol, Model.nz_state_var, Model.n_diag, Model.nobs_non_statevar);
             mexErrCheck('block_kalman_filter', err);
+        elseif DynareOptions.fast_kalman_filter
+            [LIK,lik] = kalman_filter_fast(Y,diffuse_periods+1,size(Y,2), ...
+                                           a,Pstar, ...
+                                           kalman_tol, riccati_tol, ...
+                                           DynareOptions.presample, ...
+                                           T,Q,R,H,Z,mm,pp,rr,Zflag,diffuse_periods, ...
+                                           analytic_deriv_info{:});
         else
             [LIK,lik] = kalman_filter(Y,diffuse_periods+1,size(Y,2), ...
                                       a,Pstar, ...

matlab/dynare_estimation_1.m

@@ -206,7 +206,21 @@ if isequal(options_.mode_compute,0) && isempty(options_.mode_file) && options_.m
     return
 end
 
-% Estimation of the posterior mode or likelihood mode
+%% Estimation of the posterior mode or likelihood mode
+
+% analytical derivation is not yet available for kalman_filter_fast
+if options_.analytic_derivation && options_.fast_kalman_filter
+    error(['estimation option conflict: analytic_derivation isn''t available ' ...
+           'for fast_kalman_filter'])
+end
+
+% fast kalman filter is only available with kalman_algo == 1,3
+if options_.fast_kalman_filter && ...
+        (options_.kalman_algo == 1 || options_.kalman_algo == 3)
+    error(['estimation option conflict: fast_kalman_filter is only available ' ...
+           'with kalman_algo = 1 or kalman_algo = 3'])
+end
+
 if ~isequal(options_.mode_compute,0) && ~options_.mh_posterior_mode_estimation
     %prepare settings for newrat
     if options_.mode_compute==5

matlab/global_initialization.m

@@ -407,7 +407,7 @@ options_.filtered_vars = 0;
 options_.first_obs = NaN;
 options_.nobs = NaN;
 options_.kalman_algo = 0;
-options_.fast_kalman = 0;
+options_.fast_kalman_filter = 0;
 options_.kalman_tol = 1e-10;
 options_.diffuse_kalman_tol = 1e-6;
 options_.use_univariate_filters_if_singularity_is_detected = 1;

preprocessor/DynareBison.yy

@@ -101,7 +101,7 @@ class ParsingDriver;
 %token IDENTIFICATION INF_CONSTANT INITVAL INITVAL_FILE BOUNDS JSCALE INIT INFILE INVARS
 %token <string_val> INT_NUMBER
 %token INV_GAMMA_PDF INV_GAMMA1_PDF INV_GAMMA2_PDF IRF IRF_SHOCKS IRF_PLOT_THRESHOLD IRF_CALIBRATION
-%token KALMAN_ALGO KALMAN_TOL DIFFUSE_KALMAN_TOL SUBSAMPLES OPTIONS TOLF
+%token FAST_KALMAN_FILTER KALMAN_ALGO KALMAN_TOL DIFFUSE_KALMAN_TOL SUBSAMPLES OPTIONS TOLF
 %token LAPLACE LIK_ALGO LIK_INIT LINEAR LOAD_IDENT_FILES LOAD_MH_FILE LOAD_PARAMS_AND_STEADY_STATE LOGLINEAR LOGDATA LYAPUNOV LINEAR_APPROXIMATION
 %token LYAPUNOV_FIXED_POINT_TOL LYAPUNOV_DOUBLING_TOL LYAPUNOV_SQUARE_ROOT_SOLVER_TOL LOG_DEFLATOR LOG_TREND_VAR LOG_GROWTH_FACTOR MARKOWITZ MARGINAL_DENSITY MAX MAXIT
 %token MFS MH_CONF_SIG MH_DROP MH_INIT_SCALE MH_JSCALE MH_MODE MH_NBLOCKS MH_REPLIC MH_RECOVER POSTERIOR_MAX_SUBSAMPLE_DRAWS MIN MINIMAL_SOLVING_PERIODS
@@ -1698,6 +1698,7 @@ estimation_options : o_datafile
                    | o_moments_varendo
                    | o_contemporaneous_correlation
                    | o_filtered_vars
+                   | o_fast_kalman_filter
                    | o_kalman_algo
                    | o_kalman_tol
 		   | o_diffuse_kalman_tol
@@ -2746,6 +2747,7 @@ o_moments_varendo : MOMENTS_VARENDO { driver.option_num("moments_varendo", "1");
 o_contemporaneous_correlation : CONTEMPORANEOUS_CORRELATION { driver.option_num("contemporaneous_correlation", "1"); };
 o_filtered_vars : FILTERED_VARS { driver.option_num("filtered_vars", "1"); };
 o_relative_irf : RELATIVE_IRF { driver.option_num("relative_irf", "1"); };
+o_fast_kalman_filter : FAST_KALMAN_FILTER  { driver.option_num("fast_kalman_filter", "1"); };
 o_kalman_algo : KALMAN_ALGO EQUAL INT_NUMBER { driver.option_num("kalman_algo", $3); };
 o_kalman_tol : KALMAN_TOL EQUAL non_negative_number { driver.option_num("kalman_tol", $3); };
 o_diffuse_kalman_tol : DIFFUSE_KALMAN_TOL EQUAL non_negative_number { driver.option_num("diffuse_kalman_tol", $3); };

preprocessor/DynareFlex.ll

@@ -295,6 +295,7 @@ DATE -?[0-9]+([YyAa]|[Mm]([1-9]|1[0-2])|[Qq][1-4]|[Ww]([1-9]{1}|[1-4][0-9]|5[0-2
 <DYNARE_STATEMENT>logdata 	{return token::LOGDATA;}
 <DYNARE_STATEMENT>nodiagnostic 	{return token::NODIAGNOSTIC;}
 <DYNARE_STATEMENT>kalman_algo 	{return token::KALMAN_ALGO;}
+<DYNARE_STATEMENT>fast_kalman_filter {return token::FAST_KALMAN_FILTER;}
 <DYNARE_STATEMENT>kalman_tol 	{return token::KALMAN_TOL;}
 <DYNARE_STATEMENT>diffuse_kalman_tol 	{return token::DIFFUSE_KALMAN_TOL;}
 <DYNARE_STATEMENT>forecast 	{return token::FORECAST;}

tests/Makefile.am

@@ -4,6 +4,7 @@ MODFILES = \
 	estimation/fs2000_initialize_from_calib.mod	\
 	estimation/fs2000_calibrated_covariance.mod \
 	estimation/fs2000_model_comparison.mod \
+	estimation/fs2000_fast.mod \
 	estimation/MH_recover/fs2000_recover.mod \
 	estimation/t_proposal/fs2000_student.mod \
 	estimation/TaRB/fs2000_tarb.mod \