Mean_Variance_Optimizer

A class called Mean_Variance_Optimizer was designed to implement the function of connecting to the Binance API interface to obtain perpetual contract data and calculate the mean-variance optimal portfolio weights

Class Attributes

contract_list: list[str] A list contains all the contract we want to use to construct the portofolio e.g. ['BTCUSD_PERP', 'ETHSD_PERP', 'LTCSD_PERP'] interval: str The frequency of the data e.g. '1m' for 1 minute; '1h' for 1 hour; '1d' for 1 day; '1w' for 1 week; '1M' for 1 month start_time: str The start time of our data e.g. "2023-09-01" end_time: str The end time of our data e.g. "2023-09-02"

How to run

Download this repository then enter the project directory in the terminal and run this code

pipenv install -r requirements.txt
pipenv run python main.py

The plot will pop out and the optimal weights will show in the terminal

Class method

get_data(symbol, interval, start_time, end_time)

Using requests to get the data of specific contract {symbol} with frequency of {interval} between {start_time} and {end_time} Only return the return rate

clean_data(contract_list, interval, start_time, end_time)

For every contract in the {contract_list} call the {get_data} and merge the result Result for clean_data(['BTCUSD_PERP', 'ETHSD_PERP', 'LTCSD_PERP'], '1h', "2023-09-01", "2023-09-02") is

trans_date(date_str)

Convert the date string {date_str} to a millisecond timestamp

trans_time(timestamp)

Convert the millisecond timestamp {timestamp} to a date string

expected_return(all_data)

all_data is the output from {clean_data} Compute the mean value for each contract and use it as expected return rate vector

covariance(all_data)

all_data is the output from {clean_data} Compute the covariance matrix for these given contracts and use it as expected covariance matrix

The expected return rate vector and covariance matrix are:

mean_variance_optimization(self, lambda_val)

Calculate the mean-variance optimal combination weights by solving the optimization problem

Minimize: λ * x^TΣx Maximize: x^Tμ Subject to: Σxi = 1

Plot the effective frontier variance curve and return the optimal weights as a dictionary

{'perpetual_contract_name1': weight_1, 'perpetual_contract_name2': weight_2, etc...}

Example

contract_list = ['BTCUSD_PERP', 'ETHUSD_PERP', 'LTCUSD_PERP']
interval = '1h'
start_time = '2023-09-01'
end_time = '2023-09-02'
lambda_val = 0.8

optimizer = Mean_Variance_Optimizer(contract_list, interval, start_time, end_time)
print(optimizer.mean_variance_optimization(lambda_val))

Result

The optimal weights are: {'BTCUSD_PERP': 2.4292842798086905e-07, 'ETHUSD_PERP': 0.9999996662792235, 'LTCUSD_PERP': 9.078471962833518e-08}

The the effective frontier variance curve is

Summary

From the results, we can see that the weights are all assigned to ETH, while the effective frontier curve is a straight line, which is not a good result. Personally, I think there are too few data points resulting in the expected return and covariance estimates of the sample not reflecting the real situation