前一阵子在同学的鼓动下，花了一个多月研究了股票行情的预测方法，熟悉了常见的炒股术语及技术指标，现总结如下，纯属兴趣，如果想依照本文的方法来短线操作获利，请绕道。
研究的第二步就是了解常用的股票技术指标，如macd、rsi、ar等，由于前面已经下载了所有的股票数据。那么使用下面代码就可以获得指定股票的所有历史行情：
import pandas as pdimport sqlite3 as dbcxn = db.connect('all_tushare_data.db')m= pd.read_sql("select * from gp_record where code ='%s' % id, cxn) #id是股票代码
然后就可以计算各个指标，指标的含义及计算公式可自行百度，下面实现过程仅作参考：
if 1:#OBV线，称为成交量多空比率净额法obv = m.volume*((m.close-m.low)-(m.high-m.close))/(m.high-m.low)obv = obv.fillna(0.0)obv.apply(np.cumsum)print 'obv', len(obv)if 1:#MACD指标#ema12 = m.adjust_price.copy()#以后复权价为基础ema12 = m.close.copy()ema26 = ema12.copy() for k in range(1,len(m)): ema12[k] = (ema12[k-1]*11 + ema12[k]*2)/13.0for k in range(1,len(m)): ema26[k] = (ema26[k-1]*25 + ema26[k]*2)/27.0dif = ema12-ema26dea = dif.copy()for k in range(1,len(dea)): dea[k] = (dea[k-1]*8 + dea[k]*2)/10.0macd = dif-deaprint 'macd', len(ema12), len(ema26), len(macd)if 1:#rsi指标，统计近段时间收盘涨数和跌数来判断买卖意向n=6rsi = m.change.copy()rsi.apply(np.sign)rsi = rsi.rolling(center=False,window=n).mean()print 'rsi', len(rsi)if 1:#AR BR CR 指标#第一天的计算数据没有，少一个refVa = m.open[1:].values #ArrefVb = m.close[:-1].values#BRrefVc = (2*m.close[:-1] + m.high[:-1] +m.low[:-1]).values /4#CRref = np.median([refVa, refVb, refVc])a = m.high[1:] - refb = ref - m.low[1:]c = pd.rolling_sum(a, 26)d = pd.rolling_sum(b, 26)cr = c/dprint 'ar br cr', len(cr)if 1:#kdj 指标n = 5Ln = pd.rolling_min(m.low, n)Hn = pd.rolling_max(m.high, n)rsv = (m.close-Ln)/(Hn-Ln) #跟wr指标类似para1 = para2 = 1.0/3K = rsv.copy()#以后复权价为基础K = K.fillna(0.0)for k in range(1,len(K)): K[k] = K[k-1]*(1-para1) + K[k]*para1D = K.copy()for k in range(1,len(D)): D[k] = D[k-1]*(1-para2) + D[k]*para2print 'KDJ', len(rsv), len(K), len(D)if 1:#cci 顺势指标n= 10ma = pd.rolling_mean(m.close, n)md = pd.rolling_mean((ma-m.close).abs(), n)#绝对偏差的平均值cci = ((m.close+m.high+m.low)/3 -ma)/md/0.015print 'cci', len(cci)if 1:#DMI指标 #第一天的计算数据没有，少一个n=12dm1 = pd.rolling_apply(m.high, 2, lambda d:max(d[1]-d[0],0.0))[1:]dm2 = pd.rolling_apply(m.low , 2, lambda d:max(d[0]-d[1],0.0))[1:]dm1[dm1<dm2] = 0dm2[dm2<dm1] = 0tr = (m.high-m.low).abs().values[1:], m.high[1:]-m.close[:-1].abs().values, m.low[1:]-m.close[:-1].abs().valuestr = np.minimum(np.minimum(tr[0], tr[1]) ,tr[2])dm1 = pd.rolling_mean(dm1, n) dm2 = pd.rolling_mean(dm2, n) tr = pd.rolling_mean(tr , n)di1 = dm1/tr di2 = dm2/trdx = (di1-di2).abs()/(di1+di2)adx = pd.rolling_mean(dx , n)print 'dmi', len(adx)if 1:#boll指标n=10ma = pd.rolling_mean(m.close , n)md = np.sqrt( pd.rolling_mean((m.close-ma)**2 , n))up = ma + 2*mddn = ma - 2*mdprint 'boll', len(ma), len(up), len(dn)
