Rank IC 的计算问题

2020-6-17
最近实习中研究基金经理的风格归因，需要计算因子的Rank IC。数据集是matlab的.mat文件。
主要涉及的知识点：

• .mat文件的加载
• pandas数据处理

先贴下结果：

mark一下源码：

# -*- coding: utf-8 -*-
"""
Created on Tue Jun 16 21:10:16 2020

@author: Sun
"""

import scipy.io as sio
from pandas import DataFrame as df
from pandas import Period as prd
# import pandas as pd
# import numpy as np
# from numpy import array as na

# 读取原始数据
raw = sio.loadmat('data_managers_X(1).mat')

# 从原始数据中读取索引
factors = {}
managers = list(df(raw['Company_managers_unique'])[1].apply(lambda x: str(x)[2:-2]))
date_m = list(df(raw['Date_M'].reshape(1,-1)[0])[0].apply(lambda x: prd(x)))
date_d = list(df(raw['Date_D'].reshape(1,-1)[0])[0].apply(lambda x: prd(x)))

# 从原始数据中读取因子
factors['sue_scale'] = df(raw['vOutput_stock']['sue_scale'][0,0])
factors['sue_scale'].index=date_m
factors['sue_scale'].columns=managers

# 从原始数据中读取收益率(日)
returns = {}
returns['sue_scale'] = df(raw['vOutput_nav']['ret_managers_scale'][0,0])
returns['sue_scale'].index = date_d
returns['sue_scale'].columns = managers

# 从原始数据中读取净值(日)
net_value = {}
net_value['sue_scale'] = df(raw['vOutput_nav']['value_managers_scale'][0,0])
net_value['sue_scale'].index = date_d
net_value['sue_scale'].columns = managers

return_list=df()
period_list=[]
for year in range(2010,2020):
    # 当年5-10月收益
    try:
        nav_section = net_value['sue_scale'][(returns['sue_scale'].index>='{}-05-01'.format(year))&(returns['sue_scale'].index<'{}-10-31'.format(year))]
        ret_section = df(nav_section.iloc[-1,:]/nav_section.iloc[0,:]-1).T
        # print(ret_section)
        return_list = return_list.append(ret_section)
        period_list.append('{}-10-31'.format(year))
    except Exception as e:
        print(e)
        break
    # 当年11月-次年4月收益
    try:
        next_year = year + 1
        nav_section = net_value['sue_scale'][(returns['sue_scale'].index>='{}-11-01'.format(year))&(returns['sue_scale'].index<prd('{}-04-30'.format(next_year)))]
        ret_section = df(nav_section.iloc[-1,:]/nav_section.iloc[0,:]-1).T
        # print(ret_section)
        return_list = return_list.append(ret_section)
        period_list.append('{}-04-30'.format(next_year))
    except Exception as e:
        print(e)
        break
# 得到半年为单位的收益率
return_list.index = period_list

# 对收益率进行排名
return_rank = return_list.rank(method='min',axis=1)

# 计算因子的排名，在此之前，先把2010年4月以来的数据筛选出来
factors['sue_scale'] = factors['sue_scale'][(factors['sue_scale'].index.month.isin([4,11]))&(factors['sue_scale'].index.year>=2010)]
factors['sue_scale'].fillna(.0, inplace=True)
factors_rank = factors['sue_scale'].rank(method='min',axis=1)

# 计算rank ic
rank_ic = {}
rank_ic['sue_scale'] = [factors_rank.iloc[i,:].corr(return_rank.iloc[i+1,:]) for i in range(len(factors_rank)-2)]