# 9.2 GMVP · Global Minimum Variance Portfolio (GMVP) > 来源：https://uqer.io/community/share/55461734f9f06c1c3d688030 ```py import pandas as pd import numpy as np start = '2011-07-01' # 回测起始时间 end = '2014-08-01' # 回测结束时间 benchmark = 'SH50' # 策略参考标准 universe = ['601398.XSHG','600028.XSHG', '601988.XSHG', '600036.XSHG','600030.XSHG','601318.XSHG', '600000.XSHG', '600019.XSHG', '600519.XSHG', '601166.XSHG'] capital_base = 100000 # 起始资金 longest_history = 40 # handle_data 函数中可以使用的历史数据最长窗口长度 refresh_rate = 10 # 调仓频率，即每 refresh_rate 个交易日执行一次 handle_data() 函数 def initialize(account): # 初始化虚拟账户状态 pass def handle_data(account): # 每个交易日的买入卖出指令 history_data = account.get_attribute_history('closePrice',40) retmatrix = [] for s in account.universe: retmatrix.append([history_data[s][i]/ history_data[s][i - 1] for i in range(1,40) ]) retmatrix = np.array(retmatrix) covmatrix = np.cov(retmatrix, y=None, rowvar=1, bias=0, ddof=None) covmatrix = np.matrix(covmatrix) # 不加这句执行矩阵求逆报错 covinv = np.linalg.inv(covmatrix) one_row = np.matrix(np.ones(len(account.universe))) one_vector = np.matrix(np.ones(len(account.universe))).transpose() up = np.dot(covinv, one_vector) down = np.dot(np.dot(one_row, covinv), one_vector) weights = up/down weightsum = 0 for a in weights: weightsum += a index= 0 for s in account.universe: weigh = weights[index]/weightsum index = index + 1 amount = account.cash * weigh / account.referencePrice[s] order_to(s,amount) ```  just implement the examples in the API doc