# Differences

This shows you the differences between two versions of the page.

Both sides previous revision Previous revision Next revision | Previous revision | ||

personal:portfolio:portopt [2014/05/28 13:30] antonello [Download it] |
personal:portfolio:portopt [2018/06/18 15:11] (current) |
||
---|---|---|---|

Line 1: | Line 1: | ||

- | PortOpt [Portfolio Optimizer] is a C++ program (with Python binding) implementing the Markowitz(1952) [[http:// | + | PortOpt [Portfolio Optimizer] is an open-source (LGPLed) C++ program (with Python binding) implementing the Markowitz(1952) [[http:// |

You have to provide PortOpt (in text files or - if you use the api- using your own code) the variance/ | You have to provide PortOpt (in text files or - if you use the api- using your own code) the variance/ | ||

- | It returns the vector of assets' | + | It returns the vector of assets' |

In order to minimise the variance it internally uses [[http:// | In order to minimise the variance it internally uses [[http:// | ||

Line 11: | Line 11: | ||

===== Download it ===== | ===== Download it ===== | ||

- | * windows executable (command line tool!), linux (x64) executable and source code (for C++/Python) are available from [[https:// | + | Windows executable (command line tool!), linux (x64) executable and source code (for C++/Python) are available from [[https://. |

+ | | ||

+ | ===== Bugs and support ===== | ||

+ | | ||

+ | If you find a bug, request a feature or need support open a ticket or discuss it in the [[https://. | ||

===== Theorical Background ===== | ===== Theorical Background ===== | ||

Line 17: | Line 21: | ||

In portfolio theory agents attempts to maximise portfolio expected return for a given amount of portfolio risk, or equivalently to minimise risk for a given level of expected return. | In portfolio theory agents attempts to maximise portfolio expected return for a given amount of portfolio risk, or equivalently to minimise risk for a given level of expected return. | ||

- | {{ :portfolio_graph.png?nolink |Theoretical framework}} | + | {{ :portfolio_model.png?nolink |Theoretical framework}} |

The portfolio management can be portrayed graphically as in the above Figure, where the feasible set of variance-profitability combinations in enclosed by the blue curve and the B-D segment represents the efficient frontier, where no variance can be lowered at productivity' | The portfolio management can be portrayed graphically as in the above Figure, where the feasible set of variance-profitability combinations in enclosed by the blue curve and the B-D segment represents the efficient frontier, where no variance can be lowered at productivity' | ||

Line 24: | Line 28: | ||

In such case the indifference curves can be drawn like a bundle of straight lines having equation $prod = \alpha * var + \beta$, where $\alpha$ is the linear risk aversion coefficient and both $prod$ and $var$ refer to the overall portfolio' | In such case the indifference curves can be drawn like a bundle of straight lines having equation $prod = \alpha * var + \beta$, where $\alpha$ is the linear risk aversion coefficient and both $prod$ and $var$ refer to the overall portfolio' | ||

- | Point $B$ represents the point having the lowest possible portfolio variance. | + | Point $B$ represents the point having the lowest possible portfolio variance. \alpha$ risk aversion will choose however the tangent point $C$ that can be obtained by solving the following quadratic problem: |

\begin{equation} | \begin{equation} | ||

Line 50: | Line 54: | ||

\end{equation} | \end{equation} | ||

- | where $x_i$ is the share of the asset $i$, $p_i$ is its productivity and hence $\sum_i {x_i p_i}$ is the overall portfolio productivity and $\sum_i { \sum_j { x_i x_j \sigma_{i, | + | where $x_i$ is the share of the asset $i$, $p_i$ is its productivity, $\sigma_{i,and hence $\sum_i {x_i p_i}$ is the overall portfolio productivity and $\sum_i { \sum_j { x_i x_j \sigma_{i,is its variance. |

As the only quadratic term arises when $i=j$ and $\sigma_{i, | As the only quadratic term arises when $i=j$ and $\sigma_{i, | ||

Line 77: | Line 81: | ||

g++ -std=c++0x -O -shared -Wl, | g++ -std=c++0x -O -shared -Wl, | ||

(then please refer to the python example for usage) | (then please refer to the python example for usage) | ||

+ | |||

+ | If you want to change the output library name (e.g. you want to create _portopt_p3.so for python3 alongside _portopt.so for python2), do it in the %module variable of portopt.i and in the -soname and -o options of the linking command (and don't forget to use the right python included directory in the compilation command).\\ | ||

+ | You can then load the correct module in your script with something like: | ||

+ | import sys | ||

+ | if sys.version_info < (3, 0): | ||

+ | import portopt | ||

+ | else: | ||

+ | import portopt_p3 as portopt | ||

===== Usage ===== | ===== Usage ===== | ||

+ | |||

+ | :!: Please notice that the API changed from version 1.1, with the introduction of the '' | ||

+ | |||

+ | |||

== Linux == | == Linux == | ||

./portopt [options] | ./portopt [options] | ||

Line 89: | Line 105: | ||

Call: | Call: | ||

- | double solveport (const vector< vector < | + | double solveport (const vector< vector <, string &) |

| | ||

- | == As a lib ising Python: == | + | == As a lib using Python: == |

import portopt | import portopt | ||

- | results = portopt.solveport(var, | + | results = portopt.solveport(var,,tolerance) # tolerance is optional, default to 0.000001 |

+ | functioncost = results[0] | ||

+ | shares | ||

+ | errorcode | ||

+ | errormessage = results[3] | ||

+ | opt_mean | ||

+ | opt_var | ||

=== Options === | === Options === | ||

< | < | ||

- | -h --help Prints this help | + | -h --help Prints this help |

-v --var-file [input_var_file_name] | -v --var-file [input_var_file_name] | ||

-m --means-file [input_means_file_name] | -m --means-file [input_means_file_name] | ||

-a --alpha [alpha_coefficient] | -a --alpha [alpha_coefficient] | ||

-f --field-delimiter [field_delimiter] | -f --field-delimiter [field_delimiter] | ||

- | -s --decimal-separator [decimal-separator] '' | + | -s --decimal-separator [decimal-separator] |

+ | -t --tollerance [tolerance] | ||

</ | </ | ||

Line 109: | Line 132: | ||

* Higher the alpha, lower the agent risk aversion; | * Higher the alpha, lower the agent risk aversion; | ||

* Set a negative alpha to retrieve the portfolio with the lowest possible variance; | * Set a negative alpha to retrieve the portfolio with the lowest possible variance; | ||

- | * Set alpha to zero to retrieve the portfolio with the highest mean, indipendently from variance (solution not guaranteed to be unique); | + | * Set alpha to zero to retrieve the portfolio with the highest mean, independently from variance (solution not guaranteed to be unique); |

- | * Assets shares are returned in the x_h vector, eventual error code (0: all fine, 1: input data error, 2: no solutions) in the errorcode parameter. | + | * Assets shares are returned in the x_h vector, eventual error code (0: all fine, 1: input data error, 2: no solutions, 3: didn't solve, 4: solver internal error) in the errorcode parameter. |

+ | * Use option " | ||

| | ||

| | ||

Line 122: | Line 146: | ||

You should have received a copy of the GNU Lesser General Public License along with PortOpt. | You should have received a copy of the GNU Lesser General Public License along with PortOpt. | ||

+ | |||

+ | |||

+ | ===== Citations ===== | ||

+ | If you use this program or a derivative of it in an academic framework, please cite it!\\ | ||

+ | Please cite as: | ||

+ | * A. Dragicevic, A. Lobianco, | ||

+ | |||

+ | ===== Acknowledgements ===== | ||

+ | |||

+ | This work was supported by: | ||

+ | |||

+ | * a grant overseen by Office National des Forêts through the [[http:// | ||

+ | * the French National Research Agency through the [[http:// | ||

+ | |||