day33 Python与金融量化分析(三)

第三部分 实现简单的量化框架

框架内容：

• 开始时间、结束时间、现金、持仓数据
• 获取历史数据
• 交易函数
• 计算并绘制收益曲线
• 回测主体框架
• 计算各项指标
• 用户待写代码：初始化、每日处理函数

 

第四部分 在线平台与量化投资

本节内容：

• 第一个简单的策略（了解平台）
• 双均线策略
• 因子选股策略
• 多因子选股策略
• 小市值策略
• 海龟交易法则
• 均值回归策略
• 动量策略 
• 反转策略
• 羊驼交易法则
• PEG策略
• 鳄鱼交易法则

JoinQuant平台

• 主要框架
  • initialize
  • handle_data
  • ……
• 获取历史数据
• 交易函数
• 回测频率：
  • 按天回测
  • 按分钟回测
• 风险指标

双均线策略

• 均线：对于每一个交易日，都可以计算出前N天的移动平均值，然后把这些移动平均值连起来，成为一条线，就叫做N日移动平均线。
• 移动平均线常用线有5天、10天、30天、60天、120天和240天的指标。 5天和10天的是短线操作的参照指标，称做日均线指标； 30天和60天的是中期均线指标，称做季均线指标； 120天、240天的是长期均线指标，称做年均线指标。
• 金叉：短期均线上穿长期均线
• 死叉：短期均线下穿长期均线

# 导入函数库
import jqdata

# 初始化函数，设定基准等等
def initialize(context):
    set_benchmark('000300.XSHG')
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    
    g.security = ['601318.XSHG']
    g.p1 = 5
    g.p2 = 60
    
    
def handle_data(context, data):
    cash = context.portfolio.available_cash
    for stock in g.security:
        hist = attribute_history(stock, g.p2)
        ma60 = hist['close'].mean()
        ma5 = hist['close'][-5:].mean()
        if ma5 > ma60 and stock not in context.portfolio.positions:
            order_value(stock, cash/len(g.security))
        elif ma5 < ma60 and stock in context.portfolio.positions:
            order_target(stock, 0

因子选股策略

• 因子：
  • 标准 增长率，市值，ROE，……
• 选股策略：
  • 选取该因子最大（或最小）的N只股票持仓
• 多因子选股：如何同时考虑多个因子？

# 导入函数库
import jqdata

# 初始化函数，设定基准等等
def initialize(context):
    set_benchmark('000300.XSHG')
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    g.N= 10
    g.days = 0
    
    #获取成分股
    
    
def handle_data(context, data):
    g.days+=1
    if g.days % 30 == 0:
        g.security = get_index_stocks('000300.XSHG')
        df = get_fundamentals(query(valuation).filter(valuation.code.in_(g.security)))
        df = df.sort(columns='market_cap')
        df = df.iloc[:g.N,:]
        tohold = df['code'].values
        
        for stock in context.portfolio.positions:
            if stock not in tohold:
                order_target(stock, 0)
                
        tobuy = [stock for stock in tohold if stock not in context.portfolio.positions]
        
        if len(tobuy)>0:
            cash = context.portfolio.available_cash
            cash_every_stock = cash / len(tobuy)
            for stock in tobuy:
                order_value(stock, cash_every_stock)

均值回归理论

import jqdata
import math
import numpy as np
import pandas as pd

def initialize(context):
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    
    g.benchmark = '000300.XSHG'
    set_benchmark(g.benchmark)
    
    g.ma_days = 30
    g.stock_num = 10
    
    run_monthly(handle, 1)
    #run_monthly(handle, 11)
    
def handle(context):
    tohold = get_hold_list(context)
    for stock in context.portfolio.positions:
        if stock not in tohold:
            order_target_value(stock, 0)
    
    tobuy = [stock for stock in tohold if stock not in context.portfolio.positions]
    
    if len(tobuy)>0:
        cash = context.portfolio.available_cash
        cash_every_stock = cash / len(tobuy)
        
        for stock in tobuy:
            order_value(stock, cash_every_stock)

def get_hold_list(context):
    stock_pool = get_index_stocks(g.benchmark)
    stock_score = pd.Series(index=stock_pool)
    for stock in stock_pool:
        df = attribute_history(stock, g.ma_days, '1d', ['close'])
        ma = df.mean()[0]
        current_price = get_current_data()[stock].day_open
        ratio = (ma - current_price) / ma
        stock_score[stock] = ratio
    return stock_score.nlargest(g.stock_num).index.value

• 均值回归：“跌下去的迟早要涨上来”
• 均值回归的理论基于以下观测：价格的波动一般会以它的均线为中心。也就是说，当标的价格由于波动而偏离移动均线时，它将调整并重新归于均线。
• 偏离程度：(MA-P)/MA
• 策略：在每个调仓日进行（每月调一次仓）
  • 计算池内股票的N日移动均线；
  • 计算池内所有股票价格与均线的偏离度；
  • 选取偏离度最高的num_stocks支股票并进行调仓。

布林带策略

• 布林带/布林线/保利加通道（Bollinger Band）：由三条轨道线组成，其中上下两条线分别可以看成是价格的压力线和支撑线，在两条线之间是一条价格平均线。
• 计算公式：
  •  中间线=20日均线
  • up线=20日均线+N*SD(20日收盘价)
  • down线=20日均线-N*SD(20日收盘价)

import talib
#import numpy as np
#import pandas as pd

def initialize(context):
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    set_benchmark('000300.XSHG')
    
    g.security = ['600036.XSHG']#,'601328.XSHG','600196.XSHG','600010.XSHG']
    g.N = 2
    
# 初始化此策略
def handle_data(context, data):
    cash = context.portfolio.cash / len(g.security)
    for stock in g.security:
        df = attribute_history(stock, 20)
        middle = df['close'].mean()
        upper = df['close'].mean() + g.N * df['close'].std()
        lower = df['close'].mean() - g.N * df['close'].std()
        
        current_price = get_current_data()[stock].day_open
        # 当价格突破阻力线upper时，且拥有的股票数量>=0时，卖出所有股票
        if current_price >= upper and stock in context.portfolio.positions:
            order_target(stock, 0)
        # 当价格跌破支撑线lower时, 且拥有的股票数量<=0时，则全仓买入
        elif current_price <= lower  and stock not in context.portfolio.positions:
            order_value(stock, cash)

PEG策略

• 彼得·林奇：任何一家公司股票如果定价合理的话，市盈率就会与收益增长率相等。
• 每股收益（EPS）
• 股价（P）
• 市盈率（PE）= P/EPS
• 收益增长率（G）= (EPSi – EPSi-1)/ EPSi-1
• PEG = PE / G / 100
• PEG越低，代表股价被低估的可能性越大，股价会涨的可能性越大。
• PEG是一个综合指标，既考察价值，又兼顾成长性。PEG估值法适合应用于成长型的公司。
• 注意：过滤掉市盈率或收益增长率为负的情况

def initialize(context):
    set_benchmark('000300.XSHG')
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    
    g.security = get_index_stocks('000300.XSHG')
    g.days = 0
    g.N = 10
    
def handle_data(context, data):
    g.days+=1
    if g.days%30!=0:
        return
    df = get_fundamentals(query(valuation.code, valuation.pe_ratio, indicator.inc_net_profit_year_on_year).filter(valuation.code.in_(g.security)))
    df = df[(df['pe_ratio']>0)&(df['inc_net_profit_year_on_year']>0)]
    df['PEG'] = df['pe_ratio']/df['inc_net_profit_year_on_year']/100
    df = df.sort(columns='PEG')[:g.N]
    tohold = df['code'].values
    
    for stock in context.portfolio.positions:
        if stock not in tohold:
            order_target_value(stock, 0)
    
    tobuy = [stock for stock in tohold if stock not in context.portfolio.positions]
    
    if len(tobuy)>0:
        print('Buying')
        cash = context.portfolio.available_cash
        cash_every_stock = cash / len(tobuy)
        
        for stock in tobuy:
            order_value(stock, cash_every_stock)

羊驼交易法则

• 起始时随机买入N只股票，每天卖掉收益率最差的M只，再随机买入剩余股票池的M只。

import jqdata
import pandas as pd

def initialize(context):
    set_benchmark('000300.XSHG')
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    
    g.security = get_index_stocks('000300.XSHG')
    g.period = 30
    g.N = 10
    g.change = 1
    g.init = True
    
    stocks = get_sorted_stocks(context, g.security)[:g.N]
    cash = context.portfolio.available_cash * 0.9 / len(stocks)
    for stock in stocks:
        order_value(stock, cash)
    run_monthly(handle, 2)
        
    
def get_sorted_stocks(context, stocks):
    df = history(g.period, field='close', security_list=stocks).T
    df['ret'] = (df.iloc[:,-1] - df.iloc[:,0]) / df.iloc[:,0]
    df = df.sort(columns='ret', ascending=False)
    return df.index.values
    
def handle(context):
    if g.init:
        g.init = False
        return
    stocks = get_sorted_stocks(context, context.portfolio.positions.keys())
    
    for stock in stocks[-g.change:]:
        order_target(stock, 0)
    
    stocks = get_sorted_stocks(context, g.security)
    
    for stock in stocks:
        if len(context.portfolio.positions) >= g.N:
            break
        if stock not in context.portfolio.positions:
            order_value(stock, context.portfolio.available_cash * 0.9)

海龟交易法则

• 唐奇安通道：
  • 上线=Max（前N个交易日的最高价）
  • 下线=Min（前N个交易日的最低价）
  • 中线=（上线+下线）/2

 

分钟回测

• 入市：若当前价格高于过去20日的最高价，则买入一个Unit
• 加仓：若股价在上一次买入（或加仓）的基础上上涨了0.5N，则加仓一个Unit
• 止盈：当股价跌破10日内最低价时（10日唐奇安通道下沿），清空头寸
• 止损：当价格比最后一次买入价格下跌2N时，则卖出全部头寸止损（损失不会超过2%）

 

import jqdata
import math
import numpy as np
import pandas as pd
from collections import deque

def initialize(context):
    
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')

    g.security = '000060.XSHE'
    set_benchmark(g.security)
    g.in_day = 20
    g.out_day = 10
    g.today_units = 0
    g.current_units = 0
    g.N=deque(maxlen=19)
    g.current_N = 0
    g.last_buy_price = 0
    
    price = attribute_history(g.security, g.N.maxlen*2+1, '1d', ('high', 'low', 'close'))
    
    for i in range(g.N.maxlen+1, g.N.maxlen*2+1):
        li = []
        for j in range(i-19,i+1):
            a = price['high'][j]-price['low'][j]
            b = abs(price['high'][j]-price['close'][j-1])
            c = abs(price['low'][j]-price['close'][j-1])
            li.append(max(a,b,c))
        current_N = np.array(li).mean()
        g.N.append(current_N)
        
    
def before_trading_start(context):
    g.current_N = cal_N()
    g.today_units = 0

    
def handle_data(context, data):
    dt = context.current_dt
    current_price = data[g.security].price #上一分钟价格
    value = context.portfolio.total_value
    cash = context.portfolio.available_cash
    
    unit = math.floor(value * 0.01 / g.current_N)
    
        
    if g.current_units == 0:
        buy(current_price, cash, unit)
    else:
        if stop_loss(current_price):
            return
        if sell(current_price):
            return
        addin(current_price, cash, unit)
    
def cal_N():
    # if len(g.N) < g.N.maxlen:
    #     price = attribute_history(g.security, g.N.maxlen+2, '1d', ('high', 'low', 'close'))
    #     li = []
    #     for i in range(1, g.N.maxlen+2):
    #         a = price['high'][i]-price['low'][i]
    #         b = abs(price['high'][i]-price['close'][i-1])
    #         c = abs(price['low'][i]-price['close'][i-1])
    #         li.append(max(a,b,c))
    #     current_N = np.array(li).mean()
    # else:
    price = attribute_history(g.security, 2, '1d', ('high', 'low', 'close'))
    a = price['high'][1]-price['low'][1]
    b = abs(price['high'][1]-price['close'][0])
    c = abs(price['low'][1]-price['close'][0])
    current_N = (max(a,b,c) + np.array(g.N).sum())/(g.N.maxlen+1)
    g.N.append(current_N)
    return current_N
    
def buy(current_price, cash, unit):
    price = attribute_history(g.security, g.in_day, '1d', ('high',))
    if current_price > max(price['high']):
        shares = cash / current_price
        if shares >= unit:
            print("buying %d" % unit)
            o = order(g.security, unit)
            g.last_buy_price = o.price
            g.current_units += 1
            g.today_units += 1
            return True
    return False
            

def addin(current_price, cash, unit):
    if current_price >= g.last_buy_price + 0.5 * g.current_N:
        shares = cash / current_price
        if shares >= unit:
            print("adding %d" % unit)
            o = order(g.security, unit)
            g.last_buy_price = o.price
            g.current_units += 1
            g.today_units += 1
            return True
    return False
            
def sell(current_price):
    price = attribute_history(g.security, g.out_day, '1d', ('low',))
    if current_price < min(price['low']):
        print("selling")
        order_target(g.security, 0)
        g.current_units = g.today_units
        return True
    return False
        
def stop_loss(current_price):
    if current_price < g.last_buy_price - 2 * g.current_N:
        print("stop loss")
        order_target(g.security, 0)
        g.current_units = g.today_units
        return True
    return False

 

鳄鱼法则交易系统

https://www.joinquant.com/post/595?tag=new

# 导入函数库
import jqdata
import numpy as np

# 初始化函数，设定基准等等
def initialize(context):
    set_option('use_real_price', True)
    set_order_cost(OrderCost(close_tax=0.001, open_commission=0.0003, close_commission=0.0003, min_commission=5), type='stock')
    set_benchmark('000300.XSHG')
    
    g.up_price = {} #向上碎形最高价
    g.low_price = {} #向下碎形最低价
    g.up_fractal_exists = {} #判断有效向上碎形
    g.down_fractal_exists = {} #判断有效向下碎形
    g.AO_index = {} #存放连续的AO指标数据
    g.cal_AC_index = {} #计算AC指标中转存储
    g.AC_index = {} #存放连续的AC指标数据
    g.amount = {} #满仓仓位
    g.stock = get_index_stocks('000300.XSHG')
    g.buy_stock = []
    g.month = context.current_dt.month
    run_monthly(select_universe,1,'open')
    
#重置全局变量
def reset_global():
    g.up_price = {} #向上碎形最高价
    g.low_price = {} #向下碎形最低价
    g.up_fractal_exists = {} #判断有效向上碎形
    g.down_fractal_exists = {} #判断有效向下碎形
    g.AO_index = {} #存放连续的AO指标数据
    g.cal_AC_index = {} #计算AC指标中转存储
    g.AC_index = {} #存放连续的AC指标数据
    g.amount = {} #满仓仓位
    g.buy_stock = []

def initial_stock_global(stock):
    g.up_price[stock] = 0
    g.low_price[stock] = 0
    g.up_fractal_exists[stock] = False
    g.down_fractal_exists[stock] = False #判断有效向下碎形
    g.AO_index[stock] = [0] #存放连续的AO指标数据
    g.cal_AC_index[stock] = [0]  #计算AC指标中转存储
    g.AC_index[stock] = [0] #存放连续的AC指标数据
    g.amount[stock] = 0 #满仓仓位

#轮换选股后清空持仓
def reset_position(context):
    for stock in g.buy_stock:
        order_target(stock,0)
        log.info("sell %s for reset position"%stock)

#选股
def select_universe(context):
    #每三个月操作一次
    month = context.current_dt.month
    if month%6 != g.month%6:
        return
    #清空全局变量
    reset_position(context)
    reset_global()
    hist = history(30,'1d','close',g.stock,df = False)
    for stock in g.stock:
        if is_sleeping_alligator(stock,hist,20):
            g.buy_stock.append(stock)
            #初始化该股票全局变量
            initial_stock_global(stock)
    print g.buy_stock
    return None
    
#睡着的鳄鱼
def is_sleeping_alligator(stock,hist,nday):
    for i in range(nday):
        if is_struggle(stock,hist,i) == False:
            return False
    return True

#均线纠缠，BRG三线非常接近
def is_struggle(stock,hist,delta):
    blue_line = hist[stock][-21-delta:-8-delta].mean()
    red_line = hist[stock][-13-delta:-5-delta].mean()
    green_line = hist[stock][-8-delta:-3-delta].mean()
    if abs(blue_line/red_line-1)<0.02 and abs(red_line/green_line-1)<0.02:
        return True
    else:
        return False
        
#判断 向上 或 向下 碎形
def is_fractal(stock,direction):
    hist = attribute_history(stock, 5, fields=[direction])
    if direction == 'high':
        if np.all(hist.iloc[:2] < hist.iloc[2]) and np.all(hist.iloc[3:] < hist.iloc[2]):
            g.up_price[stock] = hist.iloc[2].values
            return True
    elif direction == 'low':
        if np.all(hist.iloc[:2] > hist.iloc[2]) and np.all(hist.iloc[3:] > hist.iloc[2]):
            g.low_price[stock] = hist.iloc[2].values
            return True
    return False
    
#通过比较碎形与红线位置，判断碎形是否有效
def is_effective_fractal(stock, direction):
    if is_fractal(stock,direction):
        hist = attribute_history(stock, 11)
        red_line = hist['close'][:-3].mean()
        close_price = hist['close'][-1]
        if direction == 'high':
            if close_price > red_line:
                g.up_fractal_exists[stock] = True
            else:
                g.up_fractal_exists[stock] = False
        elif direction == 'low':
            if close_price < red_line:
                g.down_fractal_exists[stock] = True
            else:
                g.down_fractal_exists[stock] = False
    

#N日内最高价格的N日线
def nday_high_point(stock,n):
    hist = history(2*n,'1d','high',[stock],df = False)[stock]
    high_point = []
    for i in range(n):
        high_point.append(max(hist[-5-i:-1-i]))
    return np.array(high_point).mean()

#N日内最低价格的N日线
def nday_low_point(stock,n):
    hist = history(2*n,'1d','low',[stock],df = False)[stock]
    low_point = []
    for i in range(n):
        low_point.append(max(hist[-5-i:-1-i]))
    return np.array(low_point).mean()

#AO=5日内（最高-最低）/2的5日移动平均-34日内（最高-最低）/2的34日移动平均
def AO_index(stock):
    g.AO_index[stock].append(nday_high_point(stock,5)/2 + nday_low_point(stock,5)/2\
                      - nday_high_point(stock,34)/2 - nday_low_point(stock,34)/2)
    return None

#AO-AO的5日平均值的5日平均
def AC_index(stock):
    AO_index(stock)
    if len(g.AO_index[stock]) >= 5:
        g.cal_AC_index[stock].append(g.AO_index[stock][-1] - np.array(g.AO_index[stock][-5:]).mean())
        if len(g.cal_AC_index[stock]) >=5:
            g.AC_index[stock].append(np.array(g.cal_AC_index[stock][-5:]).mean())

#判断序列n日上行
def is_up_going(alist,n):
    if len(alist) < n:
        return False
    for i in range(n-1):
        if alist[-(1+i)] <= alist[-(2+i)]:
            return False
    return True

#判断序列n日下行
def is_down_going(alist,n):
    if len(alist) < n:
        return False
    for i in range(n-1):
        if alist[-(1+i)] >= alist[-(2+i)]:
            return False
    return True

#碎形被突破
def active_fractal(stock,direction):
    close_price = history(1,'1d','close',[stock],df=False)[stock][0]
    if direction == 'up' and close_price > g.up_price[stock]:
        return True
    elif direction == 'down' and close_price < g.low_price[stock]:
        return True
    return False

#进场，初始仓位
def set_initial_position(stock,context):
    close_price = history(1,'1d','close',[stock],df=False)[stock][0]
    g.amount[stock] = context.portfolio.cash/close_price/len(g.buy_stock)*3
    order(stock, g.amount[stock])
    log.info("buying %s 股数为 %s"%(stock,g.amount[stock]))
    g.down_fractal_exists[stock] = False

#卖出
def sell_all_stock(stock,context):
    order_target(stock,0)
    log.info("selling %s"%stock)
    g.up_fractal_exists[stock] = False

#加仓
def adjust_position(stock,context,position):
    order(stock,g.amount[stock]*position)
    log.info("adjust position buying %s 股数为 %s"%(stock,g.amount[stock]*position))

# 计算股票前n日收益率
def security_return(days,security_code):
    hist1 = attribute_history(security_code, days + 1, '1d', 'close',df=False)
    security_returns = (hist1['close'][-1]-hist1['close'][0])/hist1['close'][0]
    return security_returns

# 止损，根据前n日收益率
def conduct_nday_stoploss(context,security_code,days,bench):
    if  security_return(days,security_code)<= bench:
        for stock in g.buy_stock:
            order_target_value(stock,0)
            log.info("Sell %s for stoploss" %stock)
        return True
    else:
        return False

# 计算股票累计收益率（从建仓至今）
def security_accumulate_return(context,data,stock):
    current_price = data[stock].price
    cost = context.portfolio.positions[stock].avg_cost
    if cost != 0:
        return (current_price-cost)/cost
    else:
        return None

# 个股止损，根据累计收益
def conduct_accumulate_stoploss(context,data,stock,bench):
    if security_accumulate_return(context,data,stock) != None\
    and security_accumulate_return(context,data,stock) < bench:
        order_target_value(stock,0)
        log.info("Sell %s for stoploss" %stock)
        return True
    else:
        return False

# 个股止盈，根据累计收益
def conduct_accumulate_stopwin(context,data,stock,bench):
    if security_accumulate_return(context,data,stock) != None\
    and security_accumulate_return(context,data,stock) > bench:
        order_target_value(stock,0)
        log.info("Sell %s for stopwin" %stock)
        return True
    else:
        return False

def handle_data(context,data):
    #大盘止损
    if conduct_nday_stoploss(context,'000300.XSHG',3,-0.03):
        return
    for stock in g.buy_stock:
        #个股止损
        if conduct_accumulate_stopwin(context,data,stock,0.3)\
        or conduct_accumulate_stoploss(context,data,stock,-0.1):
            return
        #计算AO，AC指标
        AC_index(stock)
        #空仓时，寻找机会入场
        if context.portfolio.positions[stock].amount == 0:
            #计算向上碎形
            is_effective_fractal(stock,'high')
            #有效向上碎形存在，并被突破，买入
            if g.up_fractal_exists and active_fractal(stock,'up'):
                close_price = history(5, '1d', 'close', [stock],df = False)
                if is_up_going(g.AO_index[stock],5)\
                and is_up_going(g.AC_index[stock],3)\
                and is_up_going(close_price[stock],2):
                    set_initial_position(stock,context)
        #有持仓时，加仓或离场
        else:
            #计算向下碎形
            is_effective_fractal(stock,'low')
            #出场条件1：有效向下碎形存在，并被突破，卖出
            if g.down_fractal_exists and active_fractal(stock,'down'):
                sell_all_stock(stock,context)
                return