Main Content

Generate Code to Optimize Portfolio by Using Black Litterman Approach

This example shows how to generate a MEX function and C source code from MATLAB® code that performs portfolio optimization using the Black Litterman approach.

Prerequisites

There are no prerequisites for this example.

About the hlblacklitterman Function

The hlblacklitterman.m function reads in financial information regarding a portfolio and performs portfolio optimization using the Black Litterman approach.

type hlblacklitterman
function [er, ps, w, pw, lambda, theta] = hlblacklitterman(delta, weq, sigma, tau, P, Q, Omega)%#codegen
% hlblacklitterman
%   This function performs the Black-Litterman blending of the prior
%   and the views into a new posterior estimate of the returns as
%   described in the paper by He and Litterman.
% Inputs
%   delta  - Risk tolerance from the equilibrium portfolio
%   weq    - Weights of the assets in the equilibrium portfolio
%   sigma  - Prior covariance matrix
%   tau    - Coefficiet of uncertainty in the prior estimate of the mean (pi)
%   P      - Pick matrix for the view(s)
%   Q      - Vector of view returns
%   Omega  - Matrix of variance of the views (diagonal)
% Outputs
%   Er     - Posterior estimate of the mean returns
%   w      - Unconstrained weights computed given the Posterior estimates
%            of the mean and covariance of returns.
%   lambda - A measure of the impact of each view on the posterior estimates.
%   theta  - A measure of the share of the prior and sample information in the
%            posterior precision.

% Reverse optimize and back out the equilibrium returns
% This is formula (12) page 6.
pi = weq * sigma * delta;
% We use tau * sigma many places so just compute it once
ts = tau * sigma;
% Compute posterior estimate of the mean
% This is a simplified version of formula (8) on page 4.
er = pi' + ts * P' * inv(P * ts * P' + Omega) * (Q - P * pi');
% We can also do it the long way to illustrate that d1 + d2 = I
d = inv(inv(ts) + P' * inv(Omega) * P);
d1 = d * inv(ts);
d2 = d * P' * inv(Omega) * P;
er2 = d1 * pi' + d2 * pinv(P) * Q;
% Compute posterior estimate of the uncertainty in the mean
% This is a simplified and combined version of formulas (9) and (15)
ps = ts - ts * P' * inv(P * ts * P' + Omega) * P * ts;
posteriorSigma = sigma + ps;
% Compute the share of the posterior precision from prior and views,
% then for each individual view so we can compare it with lambda
theta=zeros(1,2+size(P,1));
theta(1,1) = (trace(inv(ts) * ps) / size(ts,1));
theta(1,2) = (trace(P'*inv(Omega)*P* ps) / size(ts,1));
for i=1:size(P,1)
    theta(1,2+i) = (trace(P(i,:)'*inv(Omega(i,i))*P(i,:)* ps) / size(ts,1));
end
% Compute posterior weights based solely on changed covariance
w = (er' * inv(delta * posteriorSigma))';
% Compute posterior weights based on uncertainty in mean and covariance
pw = (pi * inv(delta * posteriorSigma))';
% Compute lambda value
% We solve for lambda from formula (17) page 7, rather than formula (18)
% just because it is less to type, and we've already computed w*.
lambda = pinv(P)' * (w'*(1+tau) - weq)';
end

% Black-Litterman example code for MatLab (hlblacklitterman.m)
% Copyright (c) Jay Walters, blacklitterman.org, 2008.
%
% Redistribution and use in source and binary forms, 
% with or without modification, are permitted provided 
% that the following conditions are met:
%
% Redistributions of source code must retain the above 
% copyright notice, this list of conditions and the following 
% disclaimer.
% 
% Redistributions in binary form must reproduce the above 
% copyright notice, this list of conditions and the following 
% disclaimer in the documentation and/or other materials 
% provided with the distribution.
%  
% Neither the name of blacklitterman.org nor the names of its
% contributors may be used to endorse or promote products 
% derived from this software without specific prior written
% permission.
%  
% THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND 
% CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, 
% INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 
% MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 
% DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR 
% CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
% SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 
% BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 
% SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 
% INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 
% WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 
% NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
% OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH 
% DAMAGE.
%
% This program uses the examples from the paper "The Intuition 
% Behind Black-Litterman Model  Portfolios", by He and Litterman,
% 1999.  You can find a copy of this  paper at the following url.
%     http:%papers.ssrn.com/sol3/papers.cfm?abstract_id=334304
%
% For more details on the Black-Litterman model you can also view
% "The BlackLitterman Model: A Detailed Exploration", by this author
% at the following url.
%     http:%www.blacklitterman.org/Black-Litterman.pdf
%

The %#codegen directive indicates that the MATLAB code is intended for code generation.

Generate the MEX Function for Testing

Generate a MEX function using the codegen command.

codegen hlblacklitterman -args {0, zeros(1, 7), zeros(7,7), 0, zeros(1, 7), 0, 0}
Code generation successful.

Before generating C code, you should first test the MEX function in MATLAB to ensure that it is functionally equivalent to the original MATLAB code and that no run-time errors occur. By default, codegen generates a MEX function named hlblacklitterman_mex in the current folder. This allows you to test the MATLAB code and MEX function and compare the results.

Run the MEX Function

Call the generated MEX function

testMex();
View 1
Country        P       mu      w*
Australia	     0	 4.328	 1.524
Canada   	     0	 7.576	 2.095
France   	 -29.5	 9.288	-3.948
Germany  	   100	 11.04	 35.41
Japan    	     0	 4.506	 11.05
UK       	 -70.5	 6.953	-9.462
USA      	     0	 8.069	 58.57
q        	     5
omega/tau	     0.0213
lambda   	     0.317
theta   	     0.0714
pr theta  	     0.929


View 1
Country        P       mu      w*
Australia	     0	 4.328	 1.524
Canada   	     0	 7.576	 2.095
France   	 -29.5	 9.288	-3.948
Germany  	   100	 11.04	 35.41
Japan    	     0	 4.506	 11.05
UK       	 -70.5	 6.953	-9.462
USA      	     0	 8.069	 58.57
q        	     5
omega/tau	     0.0213
lambda   	     0.317
theta   	     0.0714
pr theta  	     0.929

Execution Time - MATLAB function: 0.026728 seconds
Execution Time - MEX function   : 0.017718 seconds

Generate C Code

cfg = coder.config('lib');
codegen -config cfg hlblacklitterman  -args {0, zeros(1, 7), zeros(7,7), 0, zeros(1, 7), 0, 0}
Code generation successful.

Using codegen with the specified -config cfg option produces a standalone C library.

Inspect the Generated Code

By default, the code generated for the library is in the folder codegen/lib/hbblacklitterman/.

The files are:

dir codegen/lib/hlblacklitterman/
.                              ..                             _clang-format                  buildInfo.mat                  codeInfo.mat                   codedescriptor.dmr             compileInfo.mat                examples                       hlblacklitterman.a             hlblacklitterman.c             hlblacklitterman.h             hlblacklitterman.o             hlblacklitterman_data.h        hlblacklitterman_initialize.c  hlblacklitterman_initialize.h  hlblacklitterman_initialize.o  hlblacklitterman_rtw.mk        hlblacklitterman_terminate.c   hlblacklitterman_terminate.h   hlblacklitterman_terminate.o   hlblacklitterman_types.h       interface                      inv.c                          inv.h                          inv.o                          mtimes.c                       mtimes.h                       mtimes.o                       pinv.c                         pinv.h                         pinv.o                         rtGetInf.c                     rtGetInf.h                     rtGetInf.o                     rtGetNaN.c                     rtGetNaN.h                     rtGetNaN.o                     rt_nonfinite.c                 rt_nonfinite.h                 rt_nonfinite.o                 rtw_proj.tmw                   rtwtypes.h                     

Inspect the C Code for the hlblacklitterman.c Function

type codegen/lib/hlblacklitterman/hlblacklitterman.c
/*
 * Prerelease License - for engineering feedback and testing purposes
 * only. Not for sale.
 * File: hlblacklitterman.c
 *
 * MATLAB Coder version            : 24.2
 * C/C++ source code generated on  : 20-Jul-2024 12:20:30
 */

/* Include Files */
#include "hlblacklitterman.h"
#include "inv.h"
#include "mtimes.h"
#include "pinv.h"
#include "rt_nonfinite.h"
#include <emmintrin.h>

/* Function Definitions */
/*
 * hlblacklitterman
 *    This function performs the Black-Litterman blending of the prior
 *    and the views into a new posterior estimate of the returns as
 *    described in the paper by He and Litterman.
 *  Inputs
 *    delta  - Risk tolerance from the equilibrium portfolio
 *    weq    - Weights of the assets in the equilibrium portfolio
 *    sigma  - Prior covariance matrix
 *    tau    - Coefficiet of uncertainty in the prior estimate of the mean (pi)
 *    P      - Pick matrix for the view(s)
 *    Q      - Vector of view returns
 *    Omega  - Matrix of variance of the views (diagonal)
 *  Outputs
 *    Er     - Posterior estimate of the mean returns
 *    w      - Unconstrained weights computed given the Posterior estimates
 *             of the mean and covariance of returns.
 *    lambda - A measure of the impact of each view on the posterior estimates.
 *    theta  - A measure of the share of the prior and sample information in the
 *             posterior precision.
 *
 * Arguments    : double delta
 *                const double weq[7]
 *                const double sigma[49]
 *                double tau
 *                const double P[7]
 *                double Q
 *                double Omega
 *                double er[7]
 *                double ps[49]
 *                double w[7]
 *                double pw[7]
 *                double *lambda
 *                double theta[3]
 * Return Type  : void
 */
void hlblacklitterman(double delta, const double weq[7], const double sigma[49],
                      double tau, const double P[7], double Q, double Omega,
                      double er[7], double ps[49], double w[7], double pw[7],
                      double *lambda, double theta[3])
{
  __m128d r;
  __m128d r1;
  double b_er_tmp[49];
  double ts[49];
  double er_tmp[7];
  double pi[7];
  double b_P;
  double b_tmp;
  double c_P;
  double d;
  double d1;
  double d2;
  int i;
  int i1;
  int ps_tmp;
  /*  Reverse optimize and back out the equilibrium returns */
  /*  This is formula (12) page 6. */
  for (i = 0; i < 7; i++) {
    d = 0.0;
    for (i1 = 0; i1 < 7; i1++) {
      d += weq[i1] * sigma[i1 + 7 * i];
    }
    pi[i] = d * delta;
  }
  /*  We use tau * sigma many places so just compute it once */
  for (i = 0; i <= 46; i += 2) {
    _mm_storeu_pd(&ts[i],
                  _mm_mul_pd(_mm_set1_pd(tau), _mm_loadu_pd(&sigma[i])));
  }
  ts[48] = tau * sigma[48];
  /*  Compute posterior estimate of the mean */
  /*  This is a simplified version of formula (8) on page 4. */
  b_P = 0.0;
  c_P = 0.0;
  for (i = 0; i < 7; i++) {
    d = 0.0;
    d1 = 0.0;
    for (i1 = 0; i1 < 7; i1++) {
      d2 = P[i1];
      d += ts[i + 7 * i1] * d2;
      d1 += d2 * ts[i1 + 7 * i];
    }
    er_tmp[i] = d;
    d = P[i];
    b_P += d1 * d;
    c_P += d * pi[i];
  }
  b_tmp = 1.0 / (b_P + Omega);
  b_P = Q - c_P;
  /*  We can also do it the long way to illustrate that d1 + d2 = I */
  c_P = 1.0 / Omega;
  /*  Compute posterior estimate of the uncertainty in the mean */
  /*  This is a simplified and combined version of formulas (9) and (15) */
  r = _mm_set1_pd(b_tmp);
  for (i = 0; i < 7; i++) {
    __m128d r2;
    er[i] = pi[i] + er_tmp[i] * b_tmp * b_P;
    r1 = _mm_loadu_pd(&er_tmp[0]);
    d = P[i];
    r2 = _mm_set1_pd(d);
    _mm_storeu_pd(&b_er_tmp[7 * i], _mm_mul_pd(_mm_mul_pd(r1, r), r2));
    r1 = _mm_loadu_pd(&er_tmp[2]);
    _mm_storeu_pd(&b_er_tmp[7 * i + 2], _mm_mul_pd(_mm_mul_pd(r1, r), r2));
    r1 = _mm_loadu_pd(&er_tmp[4]);
    _mm_storeu_pd(&b_er_tmp[7 * i + 4], _mm_mul_pd(_mm_mul_pd(r1, r), r2));
    b_er_tmp[7 * i + 6] = er_tmp[6] * b_tmp * d;
  }
  for (i = 0; i < 7; i++) {
    for (i1 = 0; i1 < 7; i1++) {
      d = 0.0;
      for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
        d += b_er_tmp[i + 7 * ps_tmp] * ts[ps_tmp + 7 * i1];
      }
      ps_tmp = i + 7 * i1;
      ps[ps_tmp] = ts[ps_tmp] - d;
    }
  }
  /*  Compute the share of the posterior precision from prior and views, */
  /*  then for each individual view so we can compare it with lambda */
  inv(ts, b_er_tmp);
  for (i = 0; i < 7; i++) {
    for (i1 = 0; i1 < 7; i1++) {
      d = 0.0;
      for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
        d += b_er_tmp[i + 7 * ps_tmp] * ps[ps_tmp + 7 * i1];
      }
      ts[i + 7 * i1] = d;
    }
  }
  b_P = 0.0;
  for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
    b_P += ts[ps_tmp + 7 * ps_tmp];
  }
  theta[0] = b_P / 7.0;
  r = _mm_set1_pd(c_P);
  for (i = 0; i < 7; i++) {
    r1 = _mm_set1_pd(P[i]);
    _mm_storeu_pd(&b_er_tmp[7 * i],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[0]), r), r1));
    _mm_storeu_pd(&b_er_tmp[7 * i + 2],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[2]), r), r1));
    _mm_storeu_pd(&b_er_tmp[7 * i + 4],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[4]), r), r1));
    b_er_tmp[7 * i + 6] = P[6] * c_P * P[i];
  }
  for (i = 0; i < 7; i++) {
    for (i1 = 0; i1 < 7; i1++) {
      d = 0.0;
      for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
        d += b_er_tmp[i + 7 * ps_tmp] * ps[ps_tmp + 7 * i1];
      }
      ts[i + 7 * i1] = d;
    }
  }
  b_P = 0.0;
  for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
    b_P += ts[ps_tmp + 7 * ps_tmp];
  }
  theta[1] = b_P / 7.0;
  r = _mm_set1_pd(c_P);
  for (i = 0; i < 7; i++) {
    r1 = _mm_set1_pd(P[i]);
    _mm_storeu_pd(&b_er_tmp[7 * i],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[0]), r), r1));
    _mm_storeu_pd(&b_er_tmp[7 * i + 2],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[2]), r), r1));
    _mm_storeu_pd(&b_er_tmp[7 * i + 4],
                  _mm_mul_pd(_mm_mul_pd(_mm_loadu_pd(&P[4]), r), r1));
    b_er_tmp[7 * i + 6] = P[6] * c_P * P[i];
  }
  for (i = 0; i < 7; i++) {
    for (i1 = 0; i1 < 7; i1++) {
      d = 0.0;
      for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
        d += b_er_tmp[i + 7 * ps_tmp] * ps[ps_tmp + 7 * i1];
      }
      ts[i + 7 * i1] = d;
    }
  }
  b_P = 0.0;
  for (ps_tmp = 0; ps_tmp < 7; ps_tmp++) {
    b_P += ts[ps_tmp + 7 * ps_tmp];
  }
  theta[2] = b_P / 7.0;
  /*  Compute posterior weights based solely on changed covariance */
  for (i = 0; i <= 46; i += 2) {
    r = _mm_loadu_pd(&ps[i]);
    _mm_storeu_pd(
        &b_er_tmp[i],
        _mm_mul_pd(_mm_set1_pd(delta), _mm_add_pd(_mm_loadu_pd(&sigma[i]), r)));
  }
  b_er_tmp[48] = delta * (sigma[48] + ps[48]);
  inv(b_er_tmp, ts);
  /*  Compute posterior weights based on uncertainty in mean and covariance */
  /*  Compute lambda value */
  /*  We solve for lambda from formula (17) page 7, rather than formula (18) */
  /*  just because it is less to type, and we've already computed w*. */
  for (i = 0; i < 7; i++) {
    d = 0.0;
    d1 = 0.0;
    d2 = 0.0;
    for (i1 = 0; i1 < 7; i1++) {
      b_P = ts[i1 + 7 * i];
      b_tmp = er[i1] * b_P;
      d += b_tmp;
      d1 += pi[i1] * b_P;
      d2 += b_tmp;
    }
    w[i] = d;
    pw[i] = d1;
    er_tmp[i] = d2 * (tau + 1.0) - weq[i];
  }
  pinv(P, pi);
  *lambda = mtimes(pi, er_tmp);
}

/*
 * File trailer for hlblacklitterman.c
 *
 * [EOF]
 */