5.5 钟摆理论 · 钟摆理论的简单实现——完美躲过股灾和精准抄底

今天给大家简单介绍一种稳健的投资体系——钟摆理论的量化模型实现。这里要感谢@进化论一平 的雪球分享：http://xueqiu.com/8510627167/29759691

其中的核心思想是： （1）从买入操作来看：通过找到有效价格区间，要求在价格低估时买入，但这同时又要满足另一个条件，那就是趋势必须向上。二者缺一不可。 （2）从卖出操作来看：必须顺势而为，在价格超过有效价格区间以后，如果趋势不变，不要急于卖出，直到趋势改变，价格高估再卖出。二者同样缺一不可。

那么核心问题就是两个： （1）如何找到有效的价格区间？也就是，如何给出个股的估值？（2）怎样判断趋势？

为了尽量追求简单，避免太复杂的优化。我这里直接给出两个问题的简单判定方法。

（1）根据格雷厄姆的成长价值公式进行估值，并且根据A股的实际情况或者市场情绪给予一定溢价或者折价。价值=当期(正常)利润×(8.5 + 两倍的预期年增长率)，其中的当期利润使用每股收益EPS进行衡量，预期年增长率使用EGRO/5表示，其中EGRO的计算方法为5年收益关于时间（年）进行线性回归的回归系数/5年收益均值的绝对值

（2）判断趋势有两种途径结合，一种是趋势已经向上，比较简单判断方法是五日线在十日线之上（这种判断方法犯错的几率较大，读者可以自行改进），另外一种是趋势由下向上逆转，即出现明显的底部形态。关于后者，我给出的判断标准为：股价相对于近期高点大幅下跌超过downPercent（例如30%），并且收盘价在五日线十日线之下，并且收红或者收星，跌幅小于7%

接下来就是具体实现了。

defpreceding_date(date):    cal = DataAPI.TradeCalGet(exchangeCD=u"XSHG",beginDate='20110101',endDate=date,field=['calendarDate','isOpen'],pandas="1")    cal = cal[cal['isOpen']==1]    date = cal['calendarDate'].values[-2].replace('-','')return datedefduotou_5_10(date, stockList, precedingDate=True):if precedingDate:        date = preceding_date(date)    duotou = {}if stockListisNoneor len(stockList) ==0:return duotou    kLine = DataAPI.MktStockFactorsOneDayGet(tradeDate=date,secID=stockList,field=['secID','MA5','MA10'],pandas="1")    kLine = kLine.dropna()for stock, ma5, ma10in zip(kLine['secID'].values, kLine['MA5'].values, kLine['MA10'].values):if ma5 > ma10:            duotou[stock] =Trueelse:            duotou[stock] =Falsereturn duotoudefspreadRateByIntrinsicValue(account, overflow=0.0, precedingDate=True):    stock_list = account.universe    current_date = account.current_date    date = current_date.strftime('%Y%m%d')if precedingDate:        date = preceding_date(date)    eq_EPS_EGRO = DataAPI.MktStockFactorsOneDayGet(tradeDate=date,secID=stock_list,field=['secID','EPS','EGRO'],pandas="1")    eq_EPS_EGRO['Value'] = eq_EPS_EGRO['EPS']*(8.5+2*eq_EPS_EGRO['EGRO']/5)    eq_EPS_EGRO = eq_EPS_EGRO.dropna()    spread_rate = []for stock, intrinsic_valuein zip(eq_EPS_EGRO['secID'].values, eq_EPS_EGRO['Value'].values):        intrinsic_value = intrinsic_value*(1+overflow)        reference_price = account.referencePrice[stock]if reference_price >0and reference_price < intrinsic_value:            spread_rate.append((stock, (intrinsic_value-reference_price)/reference_price))return sorted(spread_rate, key=lambda k: k[-1], reverse=True)'''判断是否为底部形态，判断标准为股价相对于近期高点大幅下跌超过downPercent，并且收盘价在五日线十日线之下，并且收红或者收星，跌幅小于7%'''defisButtom(date, stockList, precedingDate=True, downPercent=0.3):    cal = DataAPI.TradeCalGet(exchangeCD=u"XSHG",beginDate='20110101',endDate=date,field=u"prevTradeDate",pandas="1")    daysAhead = cal['prevTradeDate'].values[-20].replace('-','')if precedingDate:                date = cal['prevTradeDate'].values[-1].replace('-','')    rs = {}if stockListisNoneor len(stockList) ==0:return rs    dayInfo = DataAPI.MktEqudAdjGet(secID=stockList, beginDate=daysAhead, endDate=date ,field=['secID','openPrice','closePrice','preClosePrice'],pandas="1")    dayInfo.dropna()for stockin stockList:        stockDayInfo = dayInfo[dayInfo['secID']==stock]        closePrices = stockDayInfo['closePrice'].values        ma5 = np.mean(closePrices[-5:])        ma10 = np.mean(closePrices[-10:])        closePrice = closePrices[-1]        maxClosePrice = np.max(closePrices)        openPrice = stockDayInfo['openPrice'].values[-1]        preClosePrice = stockDayInfo['preClosePrice'].values[-1]if (maxClosePrice-closePrice)/maxClosePrice > downPercentand closePrice < ma5and ma5 < ma10and (closePrice > openPriceor abs(closePrice-openPrice)/openPrice <0.02)and abs(closePrice-preClosePrice)/preClosePrice<0.07:            rs[stock] =Trueelse:            rs[stock] =Falsereturn rs

import numpyas npstart ='2013-01-01'# 回测起始时间end ='2015-10-01'# 回测结束时间benchmark ='HS300'# 策略参考标准commission = Commission(buycost=0.0008, sellcost=0.0018)# 佣金万八universe = set_universe('CYB',date=end)# Very Important Here!! 选股很重要！不要玩大烂臭！估值再低也别玩！capital_base =1000000# 起始资金freq ='d'# 策略类型，'d'表示日间策略使用日线回测，'m'表示日内策略使用分钟线回测refresh_rate =1# 调仓频率，表示执行handle_data的时间间隔，若freq = 'd'时间间隔的单位为交易日，若freq = 'm'时间间隔为分钟max_percent_of_a_stock =1.0# 单支股的最大仓位definitialize(account):# 初始化虚拟账户状态passdefhandle_data(account):# 每个交易日的买入卖出指令global max_percent_of_a_stock    buylist = []    selist = []    current_date = account.current_date    current_date = current_date.strftime('%Y%m%d')    overflow =0.15# 根据情况给予一定的溢价（例如0.1)或者折价(例如-0.1)，也可以根据市场风险程度进行动态调节（此处读者可以自行发挥）    spread_rate = dict(spreadRateByIntrinsicValue(account, overflow=overflow, precedingDate=True))    referencePortfolioValue = account.referencePortfolioValue# 获取用来计算多头形态的股票列表    stock_set_for_duotou = []    stock_set_for_duotou.extend(account.avail_secpos.keys())    stock_set_for_duotou.extend(spread_rate.keys())    stock_set_for_duotou = list(set(stock_set_for_duotou))    duotou_5_10_Map = duotou_5_10(current_date, stock_set_for_duotou, precedingDate=True)    isButtom_Map = isButtom(current_date, stock_set_for_duotou, precedingDate=True, downPercent=0.3)for stockin account.avail_secpos.keys():if stocknotin spread_rateandnot duotou_5_10_Map.get(stock,False):            selist.append(stock)for stockin selist:        sell_value = account.referencePrice[stock]*account.valid_secpos[stock]        order_to(stock,0)for stockin spread_rate.keys():if stocknotin account.valid_secpos:            buylist.append(stock)for stockin buylist:# 满足以下条件之一买入：（1）5日线在10日线之上；（2）出现底部特征if duotou_5_10_Map.get(stock,False)or isButtom_Map.get(stock,False):            buy_value = min(referencePortfolioValue*max_percent_of_a_stock, account.cash/len(buylist))if buy_value/referencePortfolioValue >=0.0001:                order_pct(stock, buy_value/referencePortfolioValue)

回测可以发现，这种投资体系能够完美躲过股灾，并且能在股灾中精准抄底获利 ：-)