[ 50ETF 期权] 1. 历史成交持仓和 PCR 数据

最后更新于:2022-04-01 21:58:45

# [ 50ETF 期权] 1. 历史成交持仓和 PCR 数据 > 来源:https://uqer.io/community/share/5604937ff9f06c597665ef34 在本文中,我们将通过量化实验室提供的数据,计算上证50ETF期权的历史成交持仓和PCR数据,并在最后利用PCR建立一个简单的择时策略 ```py from CAL.PyCAL import * import pandas as pd import numpy as np import matplotlib.pyplot as plt from matplotlib import rc rc('mathtext', default='regular') import seaborn as sns sns.set_style('white') from matplotlib import dates ``` ## 1. 期权数据接口 有关上证50ETF期权数据,量化实验室有三个接口,分别对应于不同的功能 + `DataAPI.OptGet`: 可以获取已退市和上市的所有期权的基本信息 + `DataAPI.MktOptdGet`: 拿到历史上某一天或某段时间的期权成交行情信息 + `DataAPI.MktTickRTSnapshotGet`: 此为高频数据,获取期权最新市场信息快照 在接下来对于期权的数据分析中,我们将使用这三个API提供的数据,以下为API使用示例,具体API的详情可以查看帮助文档 ```py # 使用DataAPI.OptGet,拿到已退市和上市的所有期权的基本信息 opt_info = DataAPI.OptGet(optID='', contractStatus=[u"DE", u"L"], field='', pandas="1") opt_info.head(3) ``` | | secID | optID | secShortName | tickerSymbol | exchangeCD | currencyCD | varSecID | varShortName | varTicker | varExchangeCD | ... | contMultNum | contractStatus | listDate | expYear | expMonth | expDate | lastTradeDate | exerDate | deliDate | delistDate | | --- | --- | | 0 | 510050C1503M02200.XSHG | 10000001 | 50ETF购3月2200 | 510050C1503M02200 | XSHG | CNY | 510050.XSHG | 华夏上证50ETF | 510050 | XSHG | ... | 10000 | DE | 2015-02-09 | 2015 | 3 | 2015-03-25 | 2015-03-25 | 2015-03-25 | 2015-03-26 | 2015-03-25 | | 1 | 510050C1503M02250.XSHG | 10000002 | 50ETF购3月2250 | 510050C1503M02250 | XSHG | CNY | 510050.XSHG | 华夏上证50ETF | 510050 | XSHG | ... | 10000 | DE | 2015-02-09 | 2015 | 3 | 2015-03-25 | 2015-03-25 | 2015-03-25 | 2015-03-26 | 2015-03-25 | | 2 | 510050C1503M02300.XSHG | 10000003 | 50ETF购3月2300 | 510050C1503M02300 | XSHG | CNY | 510050.XSHG | 华夏上证50ETF | 510050 | XSHG | ... | 10000 | DE | 2015-02-09 | 2015 | 3 | 2015-03-25 | 2015-03-25 | 2015-03-25 | 2015-03-26 | 2015-03-25 | ``` 3 rows × 23 columns ``` ```py #使用DataAPI.MktOptdGet,拿到历史上某一天的期权成交信息 opt_mkt = DataAPI.MktOptdGet(tradeDate='20150921', field='', pandas="1") opt_mkt.head(2) ``` | | secID | optID | ticker | secShortName | exchangeCD | tradeDate | preSettlePrice | preClosePrice | openPrice | highestPrice | lowestPrice | closePrice | settlPrice | turnoverVol | turnoverValue | openInt | | --- | --- | | 0 | 510050C1512M02100.XSHG | 10000368 | 510050C1512M02100 | 50ETF购12月2100 | XSHG | 2015-09-21 | 0.2069 | 0.1994 | 0.1955 | 0.2087 | 0.1955 | 0.2062 | 0.2062 | 21 | 43115 | 457 | | 1 | 510050P1512M01950.XSHG | 10000369 | 510050P1512M01950 | 50ETF沽12月1950 | XSHG | 2015-09-21 | 0.1037 | 0.0999 | 0.1000 | 0.1073 | 0.0905 | 0.0905 | 0.0927 | 272 | 261112 | 868 | ```py # 获取期权最新市场信息快照 opt_mkt_snapshot = DataAPI.MktOptionTickRTSnapshotGet(optionId=u"",field=u"",pandas="1") opt_mkt_snapshot[opt_mkt_snapshot.dataDate=='2015-09-22'].head(2) ``` | | optionId | timestamp | auctionPrice | auctionQty | dataDate | dataTime | highPrice | instrumentID | lastPrice | lowPrice | ... | askBook_price1 | askBook_volume1 | askBook_price2 | askBook_volume2 | askBook_price3 | askBook_volume3 | askBook_price4 | askBook_volume4 | askBook_price5 | askBook_volume5 | | --- | --- | ``` 0 rows × 37 columns ``` ## 2. 期权历史成交持仓数据图 ```py # 华夏上证50ETF收盘价数据 secID = '510050.XSHG' begin = Date(2015, 2, 9) end = Date.todaysDate() fields = ['tradeDate', 'closePrice'] etf = DataAPI.MktFunddGet(secID, beginDate=begin.toISO().replace('-', ''), endDate=end.toISO().replace('-', ''), field=fields) etf['tradeDate'] = pd.to_datetime(etf['tradeDate']) etf = etf.set_index('tradeDate') etf.tail(2) ``` | | closePrice | | --- | --- | | tradeDate | | | 2015-09-23 | 2.180 | | 2015-09-24 | 2.187 | 统计50ETF期权历史成交量和持仓量信息 ```py # 计算历史一段时间内的50ETF期权持仓量交易量数据 def getOptHistVol(beginDate, endDate): optionVarSecID = u"510050.XSHG" cal = Calendar('China.SSE') cal.addHoliday(Date(2015,9,3)) cal.addHoliday(Date(2015,9,4)) dates = cal.bizDatesList(beginDate, endDate) dates = map(Date.toDateTime, dates) columns = ['callVol', 'putVol', 'callValue', 'putValue', 'callOpenInt', 'putOpenInt', 'nearCallVol', 'nearPutVol', 'nearCallValue', 'nearPutValue', 'nearCallOpenInt', 'nearPutOpenInt', 'netVol', 'netValue', 'netOpenInt', 'volPCR', 'valuePCR', 'openIntPCR', 'nearVolPCR', 'nearValuePCR', 'nearOpenIntPCR'] hist_opt = pd.DataFrame(0.0, index=dates, columns=columns) hist_opt.index.name = 'date' # 每一个交易日数据单独计算 for date in hist_opt.index: date_str = Date.fromDateTime(date).toISO().replace('-', '') try: opt_data = DataAPI.MktOptdGet(secID=u"", tradeDate=date_str, field=u"", pandas="1") except: hist_opt = hist_opt.drop(date) continue opt_type = [] exp_date = [] for ticker in opt_data.secID.values: opt_type.append(ticker[6]) exp_date.append(ticker[7:11]) opt_data['optType'] = opt_type opt_data['expDate'] = exp_date near_exp = np.sort(opt_data.expDate.unique())[0] data = opt_data.groupby('optType') # 计算所有上市期权:看涨看跌交易量、看涨看跌交易额、看涨看跌持仓量 hist_opt['callVol'][date] = data.turnoverVol.sum()['C'] hist_opt['putVol'][date] = data.turnoverVol.sum()['P'] hist_opt['callValue'][date] = data.turnoverValue.sum()['C'] hist_opt['putValue'][date] = data.turnoverValue.sum()['P'] hist_opt['callOpenInt'][date] = data.openInt.sum()['C'] hist_opt['putOpenInt'][date] = data.openInt.sum()['P'] near_data = opt_data[opt_data.expDate == near_exp] near_data = near_data.groupby('optType') # 计算近月期权(主力合约): 看涨看跌交易量、看涨看跌交易额、看涨看跌持仓量 hist_opt['nearCallVol'][date] = near_data.turnoverVol.sum()['C'] hist_opt['nearPutVol'][date] = near_data.turnoverVol.sum()['P'] hist_opt['nearCallValue'][date] = near_data.turnoverValue.sum()['C'] hist_opt['nearPutValue'][date] = near_data.turnoverValue.sum()['P'] hist_opt['nearCallOpenInt'][date] = near_data.openInt.sum()['C'] hist_opt['nearPutOpenInt'][date] = near_data.openInt.sum()['P'] # 计算所有上市期权: 总交易量、总交易额、总持仓量 hist_opt['netVol'][date] = hist_opt['callVol'][date] + hist_opt['putVol'][date] hist_opt['netValue'][date] = hist_opt['callValue'][date] + hist_opt['putValue'][date] hist_opt['netOpenInt'][date] = hist_opt['callOpenInt'][date] + hist_opt['putOpenInt'][date] # 计算期权看跌看涨期权交易量(持仓量)的比率: # 交易量看跌看涨比率,交易额看跌看涨比率, 持仓量看跌看涨比率 # 近月期权交易量看跌看涨比率,近月期权交易额看跌看涨比率, 近月期权持仓量看跌看涨比率 # PCR = Put Call Ratio hist_opt['volPCR'][date] = round(hist_opt['putVol'][date]*1.0/hist_opt['callVol'][date], 4) hist_opt['valuePCR'][date] = round(hist_opt['putValue'][date]*1.0/hist_opt['callValue'][date], 4) hist_opt['openIntPCR'][date] = round(hist_opt['putOpenInt'][date]*1.0/hist_opt['callOpenInt'][date], 4) hist_opt['nearVolPCR'][date] = round(hist_opt['nearPutVol'][date]*1.0/hist_opt['nearCallVol'][date], 4) hist_opt['nearValuePCR'][date] = round(hist_opt['nearPutValue'][date]*1.0/hist_opt['nearCallValue'][date], 4) hist_opt['nearOpenIntPCR'][date] = round(hist_opt['nearPutOpenInt'][date]*1.0/hist_opt['nearCallOpenInt'][date], 4) return hist_opt ``` ```py begin = Date(2015, 2, 9) end = Date.todaysDate() opt_hist = getOptHistVol(begin, end) opt_hist.tail(2) ``` | | callVol | putVol | callValue | putValue | callOpenInt | putOpenInt | nearCallVol | nearPutVol | nearCallValue | nearPutValue | ... | nearPutOpenInt | netVol | netValue | netOpenInt | volPCR | valuePCR | openIntPCR | nearVolPCR | nearValuePCR | nearOpenIntPCR | | --- | --- | | date | | | | | | | | | | | | | | | | | | | | | | | 2015-09-23 | 50093 | 42910 | 37809117 | 41517121 | 269395 | 144256 | 16603 | 11494 | 6217923 | 10409963 | ... | 50576 | 93003 | 79326238 | 413651 | 0.8566 | 1.0981 | 0.5355 | 0.6923 | 1.6742 | 0.3738 | | 2015-09-24 | 29352 | 23474 | 21696859 | 22161955 | 146224 | 98350 | 19785 | 19339 | 15693989 | 14549046 | ... | 55217 | 52826 | 43858814 | 244574 | 0.7997 | 1.0214 | 0.6726 | 0.9775 | 0.9270 | 0.8012 | ``` 2 rows × 21 columns ``` ```py ## ----- 50ETF期权成交持仓数据图 ----- fig = plt.figure(figsize=(10,5)) fig.set_tight_layout(True) ax = fig.add_subplot(111) font.set_size(16) lns1 = ax.plot(opt_hist.index, opt_hist.netOpenInt, 'grey', label = u'OpenInt') lns2 = ax.plot(opt_hist.index, opt_hist.netVol, '-r', label = 'TurnoverVolume') ax2 = ax.twinx() lns3 = ax2.plot(etf.index, etf.closePrice, '-', label = 'ETF closePrice') lns = lns1+lns2+lns3 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=2) ax.grid() ax.set_xlabel(u"tradeDate") ax.set_ylabel(r"TurnoverVolume / OpenInt") ax2.set_ylabel(r"ETF closePrice") plt.title('50ETF Option TurnoverVolume / OpenInt') plt.show() ``` ![](https://docs.gechiui.com/gc-content/uploads/sites/kancloud/2016-07-30_579cbdb9814d7.png) 从上图可以看出: + 期权的交易量基本上是50ETF的反向指标 + 五月之前的疯牛中,期权日交易量处于低位 + 六月中下旬之后的暴跌时间段,期权日交易量高位运行,是不是创个新高 + 8月17日开始的这一周中,大盘风雨飘摇,50ETF探底时,期权交易量创了新高 + 目前来看,期权交易仍然活跃,但是交易量较之前数据有所回落,应该是大盘企稳的节奏 ## 3. 期权的PCR比例 期权分看跌和看涨两种,买入两种不同的期权,代表着对于后市的不同看法,因此可以引进一个量化指标,来表示对后市看衰与看涨的力量的强弱: + PCR = Put Call Ratio + PCR可以是关于成交量的PCR,可以是持仓量的PCR,也可以是成交额的PCR ```py begin = Date(2015, 2, 9) end = Date.todaysDate() opt_hist = getOptHistVol(begin, end) opt_hist.tail(2) ``` | | callVol | putVol | callValue | putValue | callOpenInt | putOpenInt | nearCallVol | nearPutVol | nearCallValue | nearPutValue | ... | nearPutOpenInt | netVol | netValue | netOpenInt | volPCR | valuePCR | openIntPCR | nearVolPCR | nearValuePCR | nearOpenIntPCR | | --- | --- | | date | | | | | | | | | | | | | | | | | | | | | | | 2015-09-23 | 50093 | 42910 | 37809117 | 41517121 | 269395 | 144256 | 16603 | 11494 | 6217923 | 10409963 | ... | 50576 | 93003 | 79326238 | 413651 | 0.8566 | 1.0981 | 0.5355 | 0.6923 | 1.6742 | 0.3738 | | 2015-09-24 | 29352 | 23474 | 21696859 | 22161955 | 146224 | 98350 | 19785 | 19339 | 15693989 | 14549046 | ... | 55217 | 52826 | 43858814 | 244574 | 0.7997 | 1.0214 | 0.6726 | 0.9775 | 0.9270 | 0.8012 | ``` 2 rows × 21 columns ``` 首先,我们来看看成交量PCR和ETF价格走势的关系 ```py ## ---------------------------------------------- ## 50ETF期权PC比例数据图 fig = plt.figure(figsize=(10,8)) fig.set_tight_layout(True) # ------ 成交量PC比例 ------ ax = fig.add_subplot(211) lns1 = ax.plot(opt_hist.index, opt_hist.volPCR, color='r', label = u'volPCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) ax.set_ylim(0, 2) hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('Volume PCR') # ------ 近月主力期权成交量PC比例 ------ ax = fig.add_subplot(212) lns1 = ax.plot(opt_hist.index, opt_hist.nearVolPCR, color='r', label = u'nearVolPCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) ax.set_ylim(0, 2) hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('Dominant Contract Volume PCR') ``` ![](https://docs.gechiui.com/gc-content/uploads/sites/kancloud/2016-07-30_579cbdb99fb51.png) 成交量数据图中,上图为全体期权的成交量PCR,下图为近月期权的成交量PCR: + 上下两图中,PCR的曲线走势基本相似,因为期权交易中,近月期权最为活跃 + ETF价格走势,和PCR走势有比较明显的负相关性 其次,我们来看看持仓量PCR和ETF价格走势的关系 ```py ## ---------------------------------------------- ## 50ETF期权PC比例数据图 fig = plt.figure(figsize=(10,8)) fig.set_tight_layout(True) # ------ 持仓量PC比例 ------ ax = fig.add_subplot(211) lns1 = ax.plot(opt_hist.index, opt_hist.openIntPCR, color='r', label = u'volPCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) ax.set_ylim(0, 2) hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('OpenInt PCR') # ------ 近月主力期权持仓量PC比例 ------ ax = fig.add_subplot(212) lns1 = ax.plot(opt_hist.index, opt_hist.nearOpenIntPCR, color='r', label = u'nearVolPCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) ax.set_ylim(0, 2) hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('Dominant Contract OpenInt PCR') ``` ![](https://docs.gechiui.com/gc-content/uploads/sites/kancloud/2016-07-30_579cbdb9c9bcb.png) 持仓量数据图中,上图为全体期权的持仓量PCR,下图为近月期权的持仓量PCR: + 上下两图中,PCR的曲线走势基本相似,因为期权交易中,近月期权最为活跃 + 实际上,近月期权十分活跃,使得近月期权的PCR系数变动往往比整体期权PCR变化更剧烈 + ETF价格走势,和PCR走势并无明显的负相关性 + 相反,ETF价格的低点,往往PCR也处于低点,这其实说明:股价大跌之后大家会选择平仓看跌期权 最后,我们来看看成交额PCR和ETF价格走势的关系 ```py ## ---------------------------------------------- ## 50ETF期权PC比例数据图 fig = plt.figure(figsize=(10,8)) fig.set_tight_layout(True) # ------ 成交额PC比例 ------ ax = fig.add_subplot(211) lns1 = ax.plot(opt_hist.index, opt_hist.valuePCR, color='r', label = u'turnoverValuePCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) #ax.set_ylim(0, 2) ax.set_yscale('log') hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('Turnover Value PCR') # ------ 近月主力期权成交额PC比例 ------ ax = fig.add_subplot(212) lns1 = ax.plot(opt_hist.index, opt_hist.nearValuePCR, color='r', label = u'turnoverValuePCR') ax2 = ax.twinx() lns2 = ax2.plot(etf.index, etf.closePrice, '-', label = 'closePrice') lns = lns1+lns2 labs = [l.get_label() for l in lns] ax.legend(lns, labs, loc=3) #ax.set_ylim(0, 2) ax.set_yscale('log') hfmt = dates.DateFormatter('%m') ax.xaxis.set_major_formatter(hfmt) ax.grid() ax.set_xlabel(u"tradeDate(Month)") ax.set_ylabel(r"PCR") ax2.set_ylabel(r"ETF ClosePrice") plt.title('Dominant Contract Turnover Value PCR') ``` ![](https://docs.gechiui.com/gc-content/uploads/sites/kancloud/2016-07-30_579cbdb9ed549.png) 成交额数据图中,上图为全体期权的成交额PCR,下图为近月期权的成交额PCR: + 上下两图中,PCR的曲线走势基本相似,因为期权交易中,近月期权最为活跃 + 实际上,近月期权PCR指数十分活跃,使得近月期权的PCR系数变动往往比整体期权PCR变化更剧烈 + 相对于成交量和持仓量PCR指标,此处的成交额PCR指标峰值往往很高,上图中近月期权的成交额PCR最大值甚至接近30,这是由于市场恐慌时候,看跌期权成交量本身就大,而交易量大往往将看跌期权的价格大幅抬高 + ETF价格走势,和PCR走势具有明显的负相关性 4. 基于期权成交额PCR的择时策略 根据成交额PCR和ETF价格走势明显的负相关性,我们建立一个非常简单的择时策略: + PCR下降时,市场情绪趋稳定,全仓买入50ETF + PCR上升时,恐慌情绪蔓延,清仓观望 ```py start = datetime(2015, 2, 9) # 回测起始时间 end = datetime(2015, 9, 21) # 回测结束时间 hist_pcr = getOptHistVol(start, end) start = datetime(2015, 2, 9) # 回测起始时间 end = datetime(2015, 9, 21) # 回测结束时间 benchmark = '510050.XSHG' # 策略参考标准 universe = ['510050.XSHG'] # 股票池 capital_base = 100000 # 起始资金 commission = Commission(0.0,0.0) refresh_rate = 1 def initialize(account): # 初始化虚拟账户状态 account.fund = universe[0] def handle_data(account): # 每个交易日的买入卖出指令 fund = account.fund # 获取回测当日的前一天日期 dt = Date.fromDateTime(account.current_date) cal = Calendar('China.IB') cal.addHoliday(Date(2015,9,3)) cal.addHoliday(Date(2015,9,4)) last_day = cal.advanceDate(dt,'-1B',BizDayConvention.Preceding) #计算出倒数第一个交易日 last_last_day = cal.advanceDate(last_day,'-1B',BizDayConvention.Preceding) #计算出倒数第二个交易日 last_day_str = last_day.strftime("%Y-%m-%d") last_last_day_str = last_last_day.strftime("%Y-%m-%d") # 计算买入卖出信号 try: # 拿取PCR数据 pcr_last = hist_pcr['valuePCR'].loc[last_day_str] pcr_last_last = hist_pcr['valuePCR'].loc[last_last_day_str] long_flag = True if (pcr_last - pcr_last_last) < 0 else False except: long_flag = True if long_flag: approximationAmount = int(account.cash / account.referencePrice[fund] / 100.0) * 100 order(fund, approximationAmount) else: # 卖出时,全仓清空 order_to(fund, 0) ``` ![](https://docs.gechiui.com/gc-content/uploads/sites/kancloud/2016-07-30_579cbdba1ff1e.jpg) 回测结果如上,需要注意的是: + 期权挂牌时间较短,回测时间短,加上期权市场参与人数少,故而回测结果可能然并卵 + 但是严格根据PCR走势买卖50ETF,还是可以比较好的避开市场大跌的风险 + 不管怎样,PCR可以作为一个择时指标来讨论 + 除了成交额PCR,还可以通过成交量、持仓量、近月成交额等等PCR建立择时策略
';