Efficient Frontier with Google Sheets#
Description: Efficient Frontier example using Google Sheets
Tags: amplpy, google-sheets, finance
Notebook author: Christian Valente <ccv@ampl.com>
Model author: Harry Markowitz
References:
Markowitz, H.M. (March 1952). “Portfolio Selection”. The Journal of Finance. 7 (1): 77–91. doi:10.2307/2975974. JSTOR 2975974.
Install needed modules and authenticate user to use google sheets#
from google.colab import auth
auth.authenticate_user()
# Install dependencies
%pip install -q amplpy gspread --upgrade
# Google Colab & Kaggle integration
from amplpy import AMPL, ampl_notebook
ampl = ampl_notebook(
modules=["cplex"], # modules to install
license_uuid="default", # license to use
) # instantiate AMPL object and register magics
Auxiliar functions#
import gspread
from google.auth import default
creds, _ = default()
gclient = gspread.authorize(creds)
def open_spreedsheet(name):
if name.startswith("https://"):
return gclient.open_by_url(name)
return gclient.open(name)
import pandas as pd
def get_worksheet_values(name):
return spreedsheet.worksheet(name).get_values(
value_render_option="UNFORMATTED_VALUE"
)
def table_to_dataframe(rows):
return pd.DataFrame(rows[1:], columns=rows[0]).set_index(rows[0][0])
def matrix_to_dataframe(rows, tr=False, nameOverride=None):
col_labels = rows[0][1:]
row_labels = [row[0] for row in rows[1:]]
def label(pair):
return pair if not tr else (pair[1], pair[0])
data = {
label((rlabel, clabel)): rows[i + 1][j + 1]
for i, rlabel in enumerate(row_labels)
for j, clabel in enumerate(col_labels)
}
df = pd.Series(data).reset_index()
name = nameOverride if nameOverride else rows[0][0]
df.columns = ["index1", "index2", name]
return df.set_index(["index1", "index2"])
Efficient Frontier Example#
Open connection to the spreadsheet and get the data, using the utility functions in ampltools to get pandas dataframes.
spreedsheet = open_spreedsheet(
"https://docs.google.com/spreadsheets/d/1d9wRk2PJgYsjiNVoKi1FcawGSVUB5mxSeQUBdEleUv4/edit?usp=sharing"
)
rows = get_worksheet_values("covariance")
# To be able to use ampl.set_data, we override the data column
# of the dataframe to the name of the parameter we will be setting (S)
covar = matrix_to_dataframe(rows, nameOverride="S")
rows = get_worksheet_values("expectedReturns")
expected = table_to_dataframe(rows)
The following is a version of Markowitz mean-variance model, that can be used to calculate the efficient frontier
%%ampl_eval
set A ordered; # assets
param S{A, A}; # cov matrix
param mu{A} default 0; # expected returns
param lb default 0;
param ub default Infinity;
param targetReturn default -Infinity;
var w{A} >= lb <= ub; # weights
var portfolioReturn >= targetReturn;
minimize portfolio_variance: sum {i in A, j in A} S[i, j] * w[i] * w[j];
maximize portfolio_return: sum{a in A} mu[a] * w[a];
s.t. def_total_weight: sum {i in A} w[i] = 1;
s.t. def_target_return: sum{a in A} mu[a] * w[a] >= targetReturn;
Using amplpy we set the data of the ampl entities using the dataframes we got from google sheets
# Set expected returns
ampl.set_data(expected, set_name="A")
# Set covariance data
ampl.set_data(covar)
Set a few options in AMPL to streamline the execution
%%ampl_eval
# Set options
option solver cplex;
option solver_msg 0;
# The following declarations are needed for the efficient frontier script
param nPoints := 20;
param minReturn;
param maxReturn;
param variances{1..nPoints+1};
param returns{1..nPoints+1};
param delta = (maxReturn-minReturn)/nPoints;
Calculate the extreme points for the efficient frontier procedure: get the minimum return by solving the min varance problem, then get the maximum return by solving the max return problem.
%%ampl_eval
# Solve the min variance problem to get the minimum return
objective portfolio_variance;
solve > NUL;
printf "Min portfolio variance: %f, return: %f\n", portfolio_variance, portfolio_return;
let minReturn := portfolio_return;
# Store the first data point in the efficient frontier
let variances[1] := portfolio_variance;
let returns[1] := minReturn;
# Solve the max return problem
objective portfolio_return;
solve > NUL;
printf "Max portfolio variance: %f, return: %f", portfolio_variance, portfolio_return;
let maxReturn := portfolio_return;
Min portfolio variance: 0.000092, return: 0.001530
Max portfolio variance: 0.000249, return: 0.002182
Now that we have the upper and lower values for the expected returns, iterate nPoints times setting the desired return at regularly increasing levels and solve the min variance problem, thus getting the points (return - variance) needed to plot the efficient frontier.
%%ampl_eval
# Switch objective to portfolio variance
objective portfolio_variance;
# Set starting point
let targetReturn := minReturn;
# Calculate the efficient frontier
for{j in 1..nPoints}
{
let targetReturn := targetReturn+delta;
solve > NUL;
printf "Return %f, variance %f\n", portfolio_return, portfolio_variance;
let returns[j+1] := portfolio_return;
let variances[j+1]:=portfolio_variance;
};
Return 0.001563, variance 0.000092
Return 0.001596, variance 0.000092
Return 0.001628, variance 0.000092
Return 0.001661, variance 0.000093
Return 0.001693, variance 0.000093
Return 0.001726, variance 0.000094
Return 0.001758, variance 0.000094
Return 0.001791, variance 0.000095
Return 0.001823, variance 0.000096
Return 0.001856, variance 0.000097
Return 0.001889, variance 0.000099
Return 0.001921, variance 0.000100
Return 0.001954, variance 0.000101
Return 0.001986, variance 0.000104
Return 0.002019, variance 0.000110
Return 0.002051, variance 0.000118
Return 0.002084, variance 0.000129
Return 0.002116, variance 0.000144
Return 0.002149, variance 0.000182
Return 0.002182, variance 0.000249
Finally, we plot the efficient frontier
import matplotlib.pyplot as plt
df = ampl.get_data("returns", "variances").toPandas()
plt.plot(df.variances * 1000, df.returns)
plt.xlabel("Variance*1000")
plt.ylabel("Expected Return")
plt.title("Efficient frontier")
plt.show()
