#ifndef __FILE_GSMUSR_H_SEEN__ #define __FILE_GSMUSR_H_SEEN__ /* ---------------------------------------------------------------------------- Copyright (C) 2008, 2009. A. Ronald Gallant Post Office Box 659 Chapel Hill NC 27514-0659 USA This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. -----------------------------------------------------------------------------*/ //#define DEBUG_GMUSR #undef DEBUG_GMUSR /* Parameter Name Meaning Plausible theta[1] delta Discount factor 0.9992 theta[2] gamma Coefficient of risk aversion 10.0 theta[3] psi Elasticity of intertemporal substitution 1.5 theta[4] mu_c; Location parameter of log consumption growth 0.0015 theta[5] rho Autoregressive parameter of lrr 0.983 theta[6] phi_e Scale parameter of lrr 0.032 theta[7] varbar Location parameter of squared volatility 0.0064^2 theta[8] nu Autoregressive parameter of squared volatility 0.988 theta[9] sig_w Scale parameter of squared volatility 0.000004 theta[10] mu_d Location parameter of log dividend growth 0.0015 theta[11] phi_d Leverage parameter on lrr 3.0 theta[12] pi_d Correlation param on consumption innovation 1.8 theta[13] phi_u Scale parameter of log dividend growth 4.1243 stats[1] br Bond returns 1.000743618 stats[2] bv Variance of bond returns 0.000024083 stats[3] sr Stock returns 1.002851 stats[4] sv Variance of stock returns 0.001771989 Parameter Name Reasonable support theta[1] delta [ 0.9, 0.99999] theta[2] gamma [ 0.25, 100.0] theta[3] psi [ 1.0001, 5.0] theta[4] mu_c [-0.02, 0.02] theta[5] rho [ 0.0, 1.0] theta[6] phi_e [ 0.0, 1.0] theta[7] varbar [ 0.0, 1.0] theta[8] nu [ 0.0, 0.9995] theta[9] sig_w [ 0.0, 1.0] theta[10] mu_d [-0.02, 0.02] theta[11] phi_d [ 0.0, 100.0] theta[12] pi_d [ 0.0, 100.0] theta[13] phi_u [ 0.0, 100.0] */ #include "libgsm.h" #include "libsnp.h" #include "snp.h" namespace gsm { class lrr_sci_mod; class stat_mod; class snp_stat_mod; typedef lrr_sci_mod sci_mod_type; typedef snp_stat_mod stat_mod_type; const INTEGER n_state = 2; //Number of state variables const INTEGER n_sci_parms = 13; //Number of sci mod parameters const INTEGER n_sci_funcs = 4; //Number of sci mod functionals const INTEGER n_stat_funcs = 9; //Number of stat mod functionals struct lrr_sci_mod_variables { INTEGER n0; // for lags and to allow transients to dissipate INTEGER n; // simulation length INTEGER n1; // for leads INTEGER nt; // nt = n0 + n + n1 scl::realmat log_consumption_growth; scl::realmat consumption_growth; scl::realmat consumption; scl::realmat long_run_risk; scl::realmat squared_volatility; scl::realmat volatility; scl::realmat log_dividend_growth; scl::realmat dividend_growth; scl::realmat dividend; scl::realmat zeta; scl::realmat Y; scl::realmat log_mrs; scl::realmat mrs; scl::realmat log_pc_ratio; scl::realmat geometric_consumption_return; scl::realmat geometric_risk_free_rate; scl::realmat log_pd_ratio; scl::realmat geometric_stock_return; lrr_sci_mod_variables() { } lrr_sci_mod_variables(INTEGER n_lag, INTEGER n_sim, INTEGER n_lead) : n0(n_lag), n(n_sim), n1(n_lead), nt(n0+n+n1), log_consumption_growth(nt,1,0.0), consumption_growth(nt,1,0.0), consumption(nt,1,0.0), long_run_risk(nt,1,0.0), squared_volatility(nt,1,0.0), volatility(nt,1,0.0), log_dividend_growth(nt,1,0.0), dividend_growth(nt,1,0.0), dividend(nt,1,0.0), zeta(nt,3,0.0), Y(nt,3,0.0), log_mrs(nt,1,0.0), mrs(nt,1,0.0), log_pc_ratio(nt,1,0.0), geometric_consumption_return(nt,1,0.0), geometric_risk_free_rate(nt,1,0.0), log_pd_ratio(nt,1,0.0), geometric_stock_return(nt,1,0.0) { } std::vector get_lrr_sci_mod_variables(scl::realmat& mv); }; struct lrr_sci_mod_parameters { REAL delta; // discount factor REAL gamma; // coefficient of risk aversion REAL psi; // elasticity of intertemporal substitution REAL mu_c; // location parameter of log consumption growth REAL rho; // autoregressive parameter of lrr REAL phi_e; // scale parameter of lrr REAL varbar; // location parameter of squared volatility REAL nu; // autoregressive parameter of squared volatility REAL sig_w; // scale parameter of squared volatility REAL mu_d; // location parameter of log dividend growth REAL phi_d; // leverage parameter on lrr REAL pi_d; // correlation parameter on consumption innovation REAL phi_u; // scale parameter of log dividend growth REAL rpsi; // convenience parameter REAL theta; // convenience parameter lrr_sci_mod_parameters() /* : delta(0.9992), // Values supplied by Kiku gamma(10.0), psi(1.5), mu_c(0.0015), rho(0.983), phi_e(0.032), varbar(pow(0.0064,2)), nu(0.988), sig_w(0.0000040), mu_d(0.0015), phi_d(3.0), pi_d(1.8), phi_u(4.1243) */ : delta(0.9992), // Kiku job market paper gamma(10.0), psi(1.5), mu_c(0.0015), rho(0.983), phi_e(0.032), varbar(pow(0.0064,2)), nu(0.988), sig_w(0.0000017), mu_d(0.0012), phi_d(2.8), pi_d(4.125), phi_u(6.2637) { if (psi <= 0.0) scl::error("Error, lrr_sci_mod_parameters, psi <= 0"); rpsi = 1.0/psi; theta = (1.0 - gamma)/(1.0 - rpsi); } void set_theta(const scl::realmat& parm) { if (parm.nrow() != n_sci_parms || parm.ncol() != 1) scl::error("Error, lrr_sci_mod, set_theta, bad dimension"); if (psi <= 0.0) scl::error("Error, lrr_sci_mod_parameters, psi <= 0"); delta = parm[1]; gamma = parm[2]; psi = parm[3]; mu_c = parm[4]; rho = parm[5]; phi_e = parm[6]; varbar = parm[7]; nu = parm[8]; sig_w = parm[9]; mu_d = parm[10]; phi_d = parm[11]; pi_d = parm[12]; phi_u = parm[13]; rpsi = 1.0/psi; theta = (1.0 - gamma)/(1.0 - rpsi); } void get_theta(scl::realmat& parm) { parm.resize(n_sci_parms,1); parm[1] = delta; parm[2] = gamma; parm[3] = psi; parm[4] = mu_c; parm[5] = rho; parm[6] = phi_e; parm[7] = varbar; parm[8] = nu; parm[9] = sig_w; parm[10] = mu_d; parm[11] = phi_d; parm[12] = pi_d; parm[13] = phi_u; } }; extern std::ostream& operator<<(std::ostream& os, const lrr_sci_mod_parameters& mp); class lrr_sci_mod : public libgsm::sci_mod_base { private: lrr_sci_mod_parameters mp; lrr_sci_mod_variables mv; bool gen_sim_cgsr(scl::realmat& sim); bool gen_sim_cgsr(scl::realmat& sim, scl::realmat& func); std::ostream& debug; bool debug_switch; INTEGER n_dat_vars; //Number of simulated variables public: lrr_sci_mod (const scl::realmat* dat_ptr, const std::vector& pfvec, const std::vector& alvec, std::ostream& detail); INTEGER len_parm() { return n_sci_parms; } INTEGER len_func() { return n_sci_funcs; } void get_parm(scl::realmat& parm) { mp.get_theta(parm); } void set_parm(const scl::realmat& parm) { mp.set_theta(parm); } bool support(const scl::realmat& parm); bool gen_sim(scl::realmat& sim, scl::realmat& func); scl::den_val prior(const scl::realmat& parm, const scl::realmat& func); scl::realmat annualize_parms(const scl::realmat& theta); scl::realmat annualize_funcs(const scl::realmat& stats); }; class stat_mod : public libgsm::stat_mod_base { private: scl::realmat b, B; // location function parameters scl::realmat p, P; // variance function parameters, the ARCH or MA part scl::realmat G; // variance function parameters, the GARCH or AR part const scl::realmat* Y; // the data, which is Y(d,n) INTEGER d; // dimension d of the model, inferred from row dimension of Y INTEGER n; // sample size n, inferred from column dimension of Y INTEGER lagu; // lags in mean function INTEGER lagR; // lags in MA-part of variance function INTEGER lagG; // lags in AR-part of variance function INTEGER lR; // dimension of vech(R), R is Cholesky factor of variance INTEGER len_sim; REAL get_eta(INTEGER i); void set_eta(INTEGER i, REAL v); public: stat_mod (const scl::realmat* dat_ptr, const std::vector& pfvec, const std::vector& alvec, std::ostream& detail); INTEGER len_parm() { return d + d*d*lagu + lR + lR*d*lagR + lagG; } INTEGER len_func() { return n_stat_funcs; } void get_parm(scl::realmat& parm); void set_parm(const scl::realmat& parm); void set_data_ptr(const scl::realmat* dat_ptr); bool gen_sim(scl::realmat& sim, scl::realmat& func); bool support(const scl::realmat& parm); scl::den_val prior(const scl::realmat& parm, const scl::realmat& func); scl::den_val logl(); }; class snp_stat_mod : public libgsm::stat_mod_base { private: scl::realmat Xlags; // First few transformed and squashed scl::realmat Ylags; // observations to get lags for gen_sim scl::realmat X; // Transformed and squashed data or simulation scl::realmat Y; // maintained by set_data_ptr for use by logl, snp::trnfrm tr; snp::snpll ll; INTEGER lsim; INTEGER spin; INTEGER lrho; INTEGER lstats; bool snp_simulate(INT_32BIT& iseed, scl::realmat& sim); public: snp_stat_mod (const scl::realmat* dat_ptr, const std::vector& pfvec, const std::vector& alvec, std::ostream& detail); INTEGER len_parm() {return lrho;} INTEGER len_func() {return lstats;} INTEGER num_obs() {return Y.get_cols();} void get_parm(scl::realmat& rho); void set_parm(const scl::realmat& rho); void set_data_ptr(const scl::realmat* dat_ptr); bool support(const scl::realmat& rho); scl::den_val prior(const scl::realmat& rho, const scl::realmat& stats); bool gen_sim(scl::realmat& sim, scl::realmat& stats); scl::den_val logl(); scl::den_val logl(scl::realmat& dlogl); REAL penalty(scl::realmat& dpenalty); }; } #endif