量化策略系列教程:17RSI策略

不知道大家最近有沒有看到Facebook的兩個AI機器人Bob和Alice對話的事情,真是給小哥嚇了一跳!Bob和Alice參加有關社交網路助手項目的機器人。就在前幾天,他倆竟然在沒有任何人干預的情況下對話了!!!爸爸真的有點怕怕!霍金曾經就警告過大家人工智慧的危險性,這真是有點可怕了,我就想到了電影終結者,如果他們的智力呈指數增長的話,那後果真是有點可怕了!

好了,有點跑題了,今天小編再給大家上個新策略RSI,大家跑跑看,看看回測結果理想不理想~證經社量化社區 - 證經社

1.策略原理及邏輯

1.1策略原理

相對強弱指數(RSI)是通過比較一段時期內的平均收盤漲數和平均收盤跌數來分析市場買沽盤的意向和實力,從而作出未來市場的走勢。

   RSI在1978年6月由Wells Wider創製的一種通過特定時期內股價的變動情況計算市場買賣力量對比,來判斷股票價格內部本質強弱、推測價格未來的變動方向的技術指標。發表在美國Commodities雜誌中(現為Future雜誌),並收錄於同年推出的New Concepts in Technical Trading Systems書中。相比起其他分析工具,RSI是其中一種較容易向大眾傳譯的計量工具,故一推出便大受歡迎。

RSI計算公式和方法

  RSI=[上昇平均數÷(上昇平均數+下跌平均數)]×100

  具體方法:

  上昇平均數是在某一段日子裡升幅數的平均而下跌平均數則是在同一段日子裡跌幅數的平均。例如我們要計算九日RSI,首先就要找出前九日內的上昇平均數及下跌平均數,舉例子如下:

  日數收市價升跌

  第一天23.70

  第二天27.90 4.20

  第三天26.50 1.40

  第四天29.60 3.10

  第五天31.10 1.50

  第六天29.40 1.70

  第七天25.50 3.90

  第八天28.90 3.40

  第九天20.50 8.40

  第十天23.20 2.80

  (1-10天之和)+15.00+15.40

(9天內)上昇平均值:15÷9=1.67 (9天內)下跌平均值:15.40÷9=1.71

  第十天上昇平均數=(4.20+3.10+1.50+3.40+2.80)/9=1.67

  第十天下降平均數=(1.40+1.70+3.90+8.40)/9=1.71

  第十天RSI=[1.67÷(1.67+1.71)]×100=49.41

  如果第十一天收市價為25.30,則

  第十一天上昇平均數=(1.67×8+2.10)÷9=1.72

  第十一天下跌平均數=1.71×8÷9=1.52

  第十一天RSI=[1.72÷(1.72+1.52)]×100=53.09

  據此可計算以後幾天的RSI。同樣,按此方法可計算其他任何日數的RSI。至於用多少日的RSI才合適。最初RSI指標提出來時是用14天,14天作為參數則成為默定值。但在實際操作中,分析者常覺得14天太長了一點,才有5天和9天之方法。

1.2策略邏輯:

RSI指標的買賣時機:

  • 當數值運行在80以上時我們稱為超買,此時一般賣出。
  • 當數值運行在20以下時我們稱為超賣,此時一般買入。

2.策略代碼

2.1配置文件【rsi_stock.ini】(提示ini配置文件,需要保存成ANSI格式)

[strategy]nusername=npassword=n;回測模式nmode=4ntd_addr=localhost:8001nstrategy_id=n;訂閱代碼注意及時更新nsubscribe_symbols=nn[backtest]nstart_time=2014-03-01 09:00:00nend_time=2016-03-18 15:00:00nn;策略初始資金ninitial_cash=1000000nn;委託量成交比率,默認=1(每個委託100%成交)ntransaction_ratio=1nn;手續費率,默認=0(不計算手續費)ncommission_ratio=0.0003nn;滑點比率,默認=0(無滑點)nslippage_ratio=0nn;行情復權模式,0=不復權,1=前復權nprice_type=1nn;基準nbench_symbol=SHSE.000903nn[para]n;數據訂閱周期nbar_type=86400nn;rsi指標參數nrsi_period=14nn;rsi指標超買、超賣參數nover_buy=85nover_sell=25nn#止盈止損n;是否固定止盈止損nis_fixation_stop=0n;是否移動止盈nis_movement_stop=1nn;移動盈利開始比率及固定盈利比率nstop_fixation_profit=0.35n;虧損比率nstop_fixation_loss=0.068nn;移動止盈比率nstop_movement_profit=0.068nn;開倉量nopen_vol=200nn;累計開倉距離當前的最大交易日n;若開倉距今超過這個日期,則認為未開過倉nopen_max_days=22nn;歷史數據長度nhist_size=18nn;開倉量n;open_vol=2000nn##############################################################n# logger settingsn##############################################################n[loggers]nkeys=rootnn[logger_root]nlevel=INFOnhandlers=console,filenn[handlers]nkeys=console,filenn[handler_file]nclass=handlers.RotatingFileHandlernargs=(rsi_stock.log,a,1000,5)nformatter=simplenn[handler_console]nclass=StreamHandlernargs = (sys.stdout,)nformatter=simplenn[formatters]nkeys = simplenn[formatter_simple]nformat=%(asctime)s - %(name)s - %(levelname)s - %(message)sndatefmt=n

2.2策略文件【rsi_stock.py】

#!/usr/bin/env pythonn# encoding: utf-8nnimport sysnimport loggingnimport logging.confignimport configparsernimport csvnimport numpy as npnimport datetimenimport talibnimport arrownfrom gmsdk import *nnEPS = 1e-6nINIT_CLOSE_PRICE = 0nnnclass RSI_STOCK(StrategyBase):n cls_config = Nonen cls_user_name = Nonen cls_password = Nonen cls_mode = Nonen cls_td_addr = Nonen cls_strategy_id = Nonen cls_subscribe_symbols = Nonen cls_stock_pool = []nn cls_backtest_start = Nonen cls_backtest_end = Nonen cls_initial_cash = 1000000n cls_transaction_ratio = 1n cls_commission_ratio = 0.0n cls_slippage_ratio = 0.0n cls_price_type = 1n cls_bench_symbol = Nonenn def __init__(self, *args, **kwargs):n super(RSI_STOCK, self).__init__(*args, **kwargs)n self.cur_date = Nonen self.dict_close = {}n self.dict_open_close_signal = {}n self.dict_entry_high_low = {}n self.dict_last_factor = {}n self.dict_open_cum_days = {}nn @classmethodn def read_ini(cls, ini_name):n """n 功能:讀取策略配置文件n """n cls.cls_config = configparser.ConfigParser()n cls.cls_config.read(ini_name)nn @classmethodn def get_strategy_conf(cls):n """n 功能:讀取策略配置文件strategy段落的值n """n if cls.cls_config is None:n returnnn cls.cls_user_name = cls.cls_config.get(strategy, username)n cls.cls_password = cls.cls_config.get(strategy, password)n cls.cls_strategy_id = cls.cls_config.get(strategy, strategy_id)n cls.cls_subscribe_symbols = cls.cls_config.get(strategy, subscribe_symbols)n cls.cls_mode = cls.cls_config.getint(strategy, mode)n cls.cls_td_addr = cls.cls_config.get(strategy, td_addr)n if len(cls.cls_subscribe_symbols) <= 0:n cls.get_subscribe_stock()n else:n subscribe_ls = cls.cls_subscribe_symbols.split(,)n for data in subscribe_ls:n index1 = data.find(.)n index2 = data.find(., index1 + 1, -1)n cls.cls_stock_pool.append(data[:index2])nn returnnn @classmethodn def get_backtest_conf(cls):n """n 功能:讀取策略配置文件backtest段落的值n """n if cls.cls_config is None:n returnnn cls.cls_backtest_start = cls.cls_config.get(backtest, start_time)n cls.cls_backtest_end = cls.cls_config.get(backtest, end_time)n cls.cls_initial_cash = cls.cls_config.getfloat(backtest, initial_cash)n cls.cls_transaction_ratio = cls.cls_config.getfloat(backtest, transaction_ratio)n cls.cls_commission_ratio = cls.cls_config.getfloat(backtest, commission_ratio)n cls.cls_slippage_ratio = cls.cls_config.getfloat(backtest, slippage_ratio)n cls.cls_price_type = cls.cls_config.getint(backtest, price_type)n cls.cls_bench_symbol = cls.cls_config.get(backtest, bench_symbol)nn returnnn @classmethodn def get_stock_pool(cls, csv_file):n """n 功能:獲取股票池中的代碼n """n csvfile = open(csv_file, r)n reader = csv.reader(csvfile)n for line in reader:n cls.cls_stock_pool.append(line[0])nn returnnn @classmethodn def get_subscribe_stock(cls):n """n 功能:獲取訂閱代碼n """n cls.get_stock_pool(stock_pool.csv)n bar_type = cls.cls_config.getint(para, bar_type)n if 86400 == bar_type:n bar_type_str = .bar. + dailyn else:n bar_type_str = .bar. + %d % cls.cls_config.getint(para, bar_type)nn cls.cls_subscribe_symbols = ,.join(data + bar_type_str for data in cls.cls_stock_pool)n returnnn def utc_strtime(self, utc_time):n """n 功能:utc轉字元串時間n """n str_time = %s % arrow.get(utc_time).to(local)n str_time.replace(T, )n str_time = str_time.replace(T, )n return str_time[:19]nn def get_para_conf(self):n """n 功能:讀取策略配置文件para(自定義參數)段落的值n """n if self.cls_config is None:n returnnn self.rsi_period = self.cls_config.getint(para, rsi_period)n self.over_buy = self.cls_config.getint(para, over_buy)n self.over_sell = self.cls_config.getint(para, over_sell)n self.hist_size = self.cls_config.getint(para, hist_size)n self.open_vol = self.cls_config.getint(para, open_vol)n self.open_max_days = self.cls_config.getint(para, open_max_days)nn self.is_fixation_stop = self.cls_config.getint(para, is_fixation_stop)n self.is_movement_stop = self.cls_config.getint(para, is_movement_stop)nn self.stop_fixation_profit = self.cls_config.getfloat(para, stop_fixation_profit)n self.stop_fixation_loss = self.cls_config.getfloat(para, stop_fixation_loss)nn self.stop_movement_profit = self.cls_config.getfloat(para, stop_movement_profit)nn returnnn def init_strategy(self):n """n 功能:策略啟動初始化操作n """n if self.cls_mode == gm.MD_MODE_PLAYBACK:n self.cur_date = self.cls_backtest_startn self.end_date = self.cls_backtest_endn else:n self.cur_date = datetime.date.today().strftime(%Y-%m-%d) + 08:00:00n self.end_date = datetime.date.today().strftime(%Y-%m-%d) + 16:00:00nn self.dict_open_close_signal = {}n self.dict_entry_high_low = {}n self.get_last_factor()n self.init_data()n self.init_entry_high_low()n returnnn def init_data(self):n """n 功能:獲取訂閱代碼的初始化數據n """n for ticker in self.cls_stock_pool:n # 初始化倉位操作信號字典n self.dict_open_close_signal.setdefault(ticker, False)nn daily_bars = self.get_last_n_dailybars(ticker, self.hist_size - 1, self.cur_date)n if len(daily_bars) <= 0:n continuenn end_daily_bars = self.get_last_n_dailybars(ticker, 1, self.end_date)n if len(end_daily_bars) <= 0:n continuenn if ticker not in self.dict_last_factor:n continuenn end_adj_factor = self.dict_last_factor[ticker]n cp_ls = [data.close * data.adj_factor / end_adj_factor for data in daily_bars]n cp_ls.reverse()nn # 留出一個空位存儲當天的一筆數據n cp_ls.append(INIT_CLOSE_PRICE)n close = np.asarray(cp_ls, dtype=np.float)nn # 存儲歷史的closen self.dict_close.setdefault(ticker, close)nn # end = time.clock()n # logging.info(init_data cost time: %f s % (end - start))nn def init_data_newday(self):n """n 功能:新的一天初始化數據n """n # 新的一天,去掉第一筆數據,並留出一個空位存儲當天的一筆數據n for key in self.dict_close:n if len(self.dict_close[key]) >= self.hist_size and abs(self.dict_close[key][-1] - INIT_CLOSE_PRICE) > EPS:n self.dict_close[key] = np.append(self.dict_close[key][1:], INIT_CLOSE_PRICE)n elif len(self.dict_close[key]) < self.hist_size and abs(self.dict_close[key][-1] - INIT_CLOSE_PRICE) > EPS:n self.dict_close[key] = np.append(self.dict_close[key][:], INIT_CLOSE_PRICE)nn # 初始化倉位操作信號字典n for key in self.dict_open_close_signal:n self.dict_open_close_signal[key] = Falsenn # 開倉後到當前的交易日天數n keys = list(self.dict_open_cum_days.keys())n for key in keys:n if self.dict_open_cum_days[key] >= self.open_max_days:n del self.dict_open_cum_days[key]n else:n self.dict_open_cum_days[key] += 1nn def get_last_factor(self):n """n 功能:獲取指定日期最新的復權因子n """n for ticker in self.cls_stock_pool:n daily_bars = self.get_last_n_dailybars(ticker, 1, self.end_date)n if daily_bars is not None and len(daily_bars) > 0:n self.dict_last_factor.setdefault(ticker, daily_bars[0].adj_factor)nn def init_entry_high_low(self):n """n 功能:獲取進場後的最高價和最低價,模擬或實盤交易啟動時載入n """n pos_list = self.get_positions()n high_list = []n low_list = []n for pos in pos_list:n symbol = pos.exchange + . + pos.sec_idn init_time = self.utc_strtime(pos.init_time)nn cur_time = datetime.datetime.now().strftime(%Y-%m-%d %H:%M:%S)nn daily_bars = self.get_dailybars(symbol, init_time, cur_time)nn high_list = [bar.high for bar in daily_bars]n low_list = [bar.low for bar in daily_bars]nn if len(high_list) > 0:n highest = np.max(high_list)n else:n highest = pos.vwapnn if len(low_list) > 0:n lowest = np.min(low_list)n else:n lowest = pos.vwapnn self.dict_entry_high_low.setdefault(symbol, [highest, lowest])nn def on_bar(self, bar):n if self.cls_mode == gm.MD_MODE_PLAYBACK:n if bar.strtime[0:10] != self.cur_date[0:10]:n self.cur_date = bar.strtime[0:10] + 08:00:00n # 新的交易日n self.init_data_newday()nn symbol = bar.exchange + . + bar.sec_idnn self.movement_stop_profit_loss(bar)n self.fixation_stop_profit_loss(bar)nn # 填充價格n if symbol in self.dict_close:n self.dict_close[symbol][-1] = bar.closenn pos = self.get_position(bar.exchange, bar.sec_id, OrderSide_Bid)nn if self.dict_open_close_signal[symbol] is False:n # 當天未有對該代碼開、平倉n if symbol in self.dict_close:n rsi_index = talib.RSI(self.dict_close[symbol], timeperiod=self.rsi_period)nn if pos is None and symbol not in self.dict_open_cum_days n and rsi_index[-1] < self.over_sell:n # 超賣n # 有開倉機會則設置已開倉的交易天數n self.dict_open_cum_days[symbol] = 0nn cash = self.get_cash()n cur_open_vol = self.open_voln if cash.available / bar.close > self.open_vol:n cur_open_vol = self.open_voln else:n cur_open_vol = int(cash.available / bar.close / 100) * 100nn if cur_open_vol == 0:n print(no available cash to buy, available cash: %.2f % cash.available)n else:n self.open_long(bar.exchange, bar.sec_id, bar.close, cur_open_vol)n self.dict_open_close_signal[symbol] = Truen logging.info(open long, symbol:%s, time:%s, price:%.2f % (symbol, bar.strtime, bar.close))n elif pos is not None and rsi_index[-1] > self.over_buy:n vol = pos.volume - pos.volume_todayn if vol > 0:n self.close_long(bar.exchange, bar.sec_id, bar.close, vol)n self.dict_open_close_signal[symbol] = Truen logging.info(close long, symbol:%s, time:%s, price:%.2f % (symbol, bar.strtime, bar.close))nn def on_order_filled(self, order):n symbol = order.exchange + . + order.sec_idn if order.position_effect == PositionEffect_CloseYesterday n and order.side == OrderSide_Bid:n pos = self.get_position(order.exchange, order.sec_id, order.side)n if pos is None and self.is_movement_stop == 1:n self.dict_entry_high_low.pop(symbol)nn def fixation_stop_profit_loss(self, bar):n """n 功能:固定止盈、止損,盈利或虧損超過了設置的比率則執行止盈、止損n """n if self.is_fixation_stop == 0:n returnnn symbol = bar.exchange + . + bar.sec_idn pos = self.get_position(bar.exchange, bar.sec_id, OrderSide_Bid)n if pos is not None:n if pos.fpnl > 0 and pos.fpnl / pos.cost >= self.stop_fixation_profit:n self.close_long(bar.exchange, bar.sec_id, 0, pos.volume - pos.volume_today)n self.dict_open_close_signal[symbol] = Truen logging.info(n fixnation stop profit: close long, symbol:%s, time:%s, price:%.2f, vwap: %s, volume:%s % (symbol,n bar.strtime,n bar.close,n pos.vwap,n pos.volume))n elif pos.fpnl < 0 and pos.fpnl / pos.cost <= -1 * self.stop_fixation_loss:n self.close_long(bar.exchange, bar.sec_id, 0, pos.volume - pos.volume_today)n self.dict_open_close_signal[symbol] = Truen logging.info(n fixnation stop loss: close long, symbol:%s, time:%s, price:%.2f, vwap:%s, volume:%s % (symbol,n bar.strtime,n bar.close,n pos.vwap,n pos.volume))nn def movement_stop_profit_loss(self, bar):n """n 功能:移動止盈, 移動止盈止損按進場後的最高價乘以設置的比率與當前價格相比,n 並且盈利比率達到設定的盈虧比率時,執行止盈n """n if self.is_movement_stop == 0:n returnnn entry_high = Nonen entry_low = Nonen pos = self.get_position(bar.exchange, bar.sec_id, OrderSide_Bid)n symbol = bar.exchange + . + bar.sec_idnn is_stop_profit = Truenn if pos is not None and pos.volume > 0:n if symbol in self.dict_entry_high_low:n if self.dict_entry_high_low[symbol][0] < bar.close:n self.dict_entry_high_low[symbol][0] = bar.closen is_stop_profit = Falsen if self.dict_entry_high_low[symbol][1] > bar.close:n self.dict_entry_high_low[symbol][1] = bar.closen [entry_high, entry_low] = self.dict_entry_high_low[symbol]nn else:n self.dict_entry_high_low.setdefault(symbol, [bar.close, bar.close])n [entry_high, entry_low] = self.dict_entry_high_low[symbol]n is_stop_profit = Falsenn if is_stop_profit:n # 移動止盈n if bar.close <= (n 1 - self.stop_movement_profit) * entry_high and pos.fpnl / pos.cost >= self.stop_fixation_profit:n if pos.volume - pos.volume_today > 0:n self.close_long(bar.exchange, bar.sec_id, 0, pos.volume - pos.volume_today)n self.dict_open_close_signal[symbol] = Truen logging.info(n movement stop profit: close long, symbol:%s, time:%s, price:%.2f, vwap:%.2f, volume:%s % (n symbol,n bar.strtime, bar.close, pos.vwap, pos.volume))nn # 止損n if pos.fpnl < 0 and pos.fpnl / pos.cost <= -1 * self.stop_fixation_loss:n self.close_long(bar.exchange, bar.sec_id, 0, pos.volume - pos.volume_today)n self.dict_open_close_signal[symbol] = Truen logging.info(n movement stop loss: close long, symbol:%s, time:%s, price:%.2f, vwap:%.2f, volume:%s % (symbol,n bar.strtime,n bar.close,n pos.vwap,n pos.volume))nnnif __name__ == __main__:n print(get_version())n logging.config.fileConfig(rsi_stock.ini)n RSI_STOCK.read_ini(rsi_stock.ini)n RSI_STOCK.get_strategy_conf()nn rsi_stock = RSI_STOCK(username=RSI_STOCK.cls_user_name,n password=RSI_STOCK.cls_password,n strategy_id=RSI_STOCK.cls_strategy_id,n subscribe_symbols=RSI_STOCK.cls_subscribe_symbols,n mode=RSI_STOCK.cls_mode,n td_addr=RSI_STOCK.cls_td_addr)nn if RSI_STOCK.cls_mode == gm.MD_MODE_PLAYBACK:n RSI_STOCK.get_backtest_conf()n ret = rsi_stock.backtest_config(start_time=RSI_STOCK.cls_backtest_start,n end_time=RSI_STOCK.cls_backtest_end,n initial_cash=RSI_STOCK.cls_initial_cash,n transaction_ratio=RSI_STOCK.cls_transaction_ratio,n commission_ratio=RSI_STOCK.cls_commission_ratio,n slippage_ratio=RSI_STOCK.cls_slippage_ratio,n price_type=RSI_STOCK.cls_price_type,n bench_symbol=RSI_STOCK.cls_bench_symbol)nn rsi_stock.get_para_conf()n rsi_stock.init_strategy()n ret = rsi_stock.run()nnprint(run result %s % ret)n

2.3[stock_pool.csv]文件

SHSE.600000nSHSE.600010nSHSE.600011nSHSE.600015nSHSE.600016nSHSE.600018nSHSE.600019nSHSE.600023nSHSE.600028nSHSE.600030nSHSE.600031nSHSE.600036nSHSE.600048nSHSE.600050nSHSE.600104nSHSE.600111nSHSE.600115nSHSE.600150nSHSE.600276nSHSE.600340nSHSE.600372nSHSE.600398nSHSE.600485nSHSE.600518nSHSE.600519nSHSE.600585nSHSE.600637nSHSE.600690nSHSE.600705nSHSE.600795nSHSE.600837nSHSE.600886nSHSE.600887nSHSE.600893nSHSE.600900nSHSE.600958nSHSE.600959nSHSE.600999nSHSE.601006nSHSE.601018nSHSE.601088nSHSE.601111nSHSE.601166nSHSE.601169nSHSE.601186nSHSE.601211nSHSE.601225nSHSE.601288nSHSE.601318nSHSE.601328nSHSE.601336nSHSE.601377nSHSE.601390nSHSE.601398nSHSE.601600nSHSE.601601nSHSE.601618nSHSE.601628nSHSE.601633nSHSE.601668nSHSE.601669nSHSE.601688nSHSE.601727nSHSE.601766nSHSE.601788nSHSE.601800nSHSE.601808nSHSE.601818nSHSE.601857nSHSE.601898nSHSE.601899nSHSE.601901nSHSE.601939nSHSE.601985nSHSE.601988nSHSE.601989nSHSE.601998nSHSE.603288nSZSE.000001nSZSE.000002nSZSE.000063nSZSE.000069nSZSE.000166nSZSE.000333nSZSE.000538nSZSE.000625nSZSE.000651nSZSE.000725nSZSE.000776nSZSE.000858nSZSE.000895nSZSE.002024nSZSE.002252nSZSE.002304nSZSE.002415nSZSE.002594nSZSE.002736nSZSE.002739nSZSE.300059nSZSE.300104n

如果想了解相關的python函數和掘金介面函數,走下方通道:

http://zjshe.cn/q/forum.php?mod=viewthread&tid=85

有不了解的可以給小編留言哦,小編會細心為大家解答~


推薦閱讀:

Python concurrent.future 使用教程及源碼初剖
機器學習2—KNN演算法(原理、實現、實例)

TAG:策略 | 量化 | Python |