예제 #1
0
def factor_leverage(cur_date):
    # 现金比率,-1列为rank,-2列为所需要的
    fa_cashtoliqdebt = om.get_fa_cashtoliqdebt(cur_date)
    # 流动比率,-1列为rank,-2列为所需要的
    fa_current = om.get_fa_current(cur_date)
    # 权益乘数,-1列为rank,-2列为所需要的
    fa_assetstoequity = om.get_fa_assetstoequity(cur_date)
    # 获取个股市值(带index)
    index_mkvalue = om.get_index_mkvalue(cur_date)

    cashration = fa_cashtoliqdebt.iloc[:, -2]  # 现金比率
    currentratio = fa_current.iloc[:, -2]  # 流动比率
    equitymultiplier = fa_assetstoequity.iloc[:, -2]
    ln_capital = np.log(index_mkvalue.iloc[:, -1])  # 市值对数
    # 入库
    stock_list = om.get_all_stock(cur_date)
    cashration_flag = fh.df_to_list(cur_date, cashration, 'cashration', stock_list)
    currentratio_flag = fh.df_to_list(cur_date, currentratio, 'currentratio', stock_list)
    equitymultiplier_flag = fh.df_to_list(cur_date, equitymultiplier, 'equitymultiplier', stock_list)
    ln_capital_flag = fh.df_to_list(cur_date, ln_capital, 'ln_capital', stock_list)

    if cashration_flag & currentratio_flag & equitymultiplier_flag & ln_capital_flag != 1:
        gl.logger.error('factor_leverage insert fail')
        return 0
    else:
        gl.logger.info('factor_leverage insert successful')
        return 1
예제 #2
0
def factor_growth(cur_date):
    # 单季度 营业收入环比增长率,-1列为rank,-2列为所需要的
    qfa_cgrsales = om.get_qfa_cgrsales(cur_date)
    # 营业收入同比增长率,-1列为rank,-2列为所需要的
    fa_yoy_or = om.get_fa_yoy_or(cur_date)
    # 单季度.净利润同比增长率(%),-1列为rank,-2列为所需要的
    qfa_yoyprofit = om.get_qfa_yoyprofit(cur_date)
    # 同比增长率-归属母公司股东的净利润-扣除非经常损益(%),-1列为rank,-2列为所需要的
    profit_deducted = om.get_profit_deducted(cur_date)
    # 同比增长率-经营活动产生的现金流量净额(%),-1列为rank,-2列为所需要的
    fa_yoyocf = om.get_fa_yoyocf(cur_date)
    # 同比增长率-归属母公司股东的净利润-扣除非经常损益(%),-1列为rank,-2列为所需要的
    profit_deducted_second = om.get_profit_deducted(cur_date - 10000)
    # 同比增长率-归属母公司股东的净利润-扣除非经常损益(%),-1列为rank,-2列为所需要的
    profit_deducted_third = om.get_profit_deducted(cur_date - 20000)
    # 同比增长率-经营活动产生的现金流量净额(%),-1列为rank,-2列为所需要的
    fa_yoyocf_second = om.get_fa_yoyocf(cur_date - 10000)
    # 同比增长率-经营活动产生的现金流量净额(%),-1列为rank,-2列为所需要的
    fa_yoyocf_third = om.get_fa_yoyocf(cur_date - 20000)
    # 营业收入同比增长率,-1列为rank,-2列为所需要的(去年的)
    fa_yoy_or_second = om.get_fa_yoy_or(cur_date - 10000)
    # 营业收入同比增长率,-1列为rank,-2列为所需要的(前年的)
    fa_yoy_or_third = om.get_fa_yoy_or(cur_date - 20000)

    sales_growth_q = qfa_cgrsales.iloc[:, -2]
    sales_growth_ttm = fa_yoy_or.iloc[:, -2]
    sales_growth_3y = (((fa_yoy_or.iloc[:, -2] / 100 + 1) * (fa_yoy_or_second.iloc[:, -2] / 100 + 1) *
                       (fa_yoy_or_third.iloc[:, -2] / 100 + 1)) ** (1 / 3) - 1) * 100
    profit_growth_q = qfa_yoyprofit.iloc[:, -2]
    profit_growth_ttm = profit_deducted.iloc[:, -2]
    profit_growth_3y = (((profit_deducted.iloc[:, -2] / 100 + 1) * (profit_deducted_second.iloc[:, -2] / 100 + 1) *
                        (profit_deducted_third.iloc[:, -2] / 100 + 1)) ** (1 / 3) - 1) * 100
    operationcashflow_growth_ttm = fa_yoyocf.iloc[:, -2]
    operationcashflow_growth_3y = (((fa_yoyocf.iloc[:, -2] / 100 + 1) * (fa_yoyocf_second.iloc[:, -2] / 100 + 1) *
                                   (fa_yoyocf_third.iloc[:, -2] / 100 + 1)) ** (1 / 3) - 1) * 100
    # 入库
    stock_list = om.get_all_stock(cur_date)
    sales_growth_q_flag = fh.df_to_list(cur_date, sales_growth_q, 'sales_growth_q', stock_list)
    sales_growth_ttm_flag = fh.df_to_list(cur_date, sales_growth_ttm, 'sales_growth_ttm', stock_list)
    sales_growth_3y_flag = fh.df_to_list(cur_date, sales_growth_3y, 'sales_growth_3y', stock_list)
    profit_growth_q_flag = fh.df_to_list(cur_date, profit_growth_q, 'profit_growth_q', stock_list)
    profit_growth_ttm_flag = fh.df_to_list(cur_date, profit_growth_ttm, 'profit_growth_ttm', stock_list)
    profit_growth_3y_flag = fh.df_to_list(cur_date, profit_growth_3y, 'profit_growth_3y', stock_list)
    operationcashflow_growth_ttm_flag = fh.df_to_list(cur_date, operationcashflow_growth_ttm,
                                                      'operationcashflow_growth_ttm', stock_list)
    operationcashflow_growth_3y_flag = fh.df_to_list(cur_date, operationcashflow_growth_3y,
                                                     'operationcashflow_growth_3y', stock_list)

    if sales_growth_q_flag & sales_growth_ttm_flag & sales_growth_3y_flag & profit_growth_q_flag & profit_growth_ttm_flag & \
            profit_growth_3y_flag & operationcashflow_growth_ttm_flag & operationcashflow_growth_3y_flag != 1:
        gl.logger.error('factor_growth insert fail')
        return 0
    else:
        gl.logger.info('factor_growth insert successful')
        return 1
예제 #3
0
def factor_close_bak(cur_date):
    close_bak = om.get_close_bak(cur_date)
    close_bak = close_bak.iloc[:, -1]
    stock_list = om.get_all_stock(cur_date)
    close_bak_flag = fh.df_to_list(cur_date, close_bak, 'close_bak', stock_list)
    if close_bak_flag != 1:
        gl.logger.error('factor_close_bak insert fail')
        return 0
    else:
        gl.logger.info('factor_close_bak insert successful')
        return 1
예제 #4
0
def factor_turnover(cur_date, date_list):
    m1, m2, m3, m6, m12 = 20, 40, 60, 120, 240
    # 此处为了获得换手率,取得成交量和流通股本
    # 获取成交量,和当日流通股本,不带index,为了格式化df成为index为日期,columns为股票代码
    dq_volume = om.get_dq_volume(cur_date, date_list)
    shr_today = om.get_shr_today(cur_date, date_list)
    # 得到近一年的最高价,最低价,并转换为index为日期,columns为股票代码
    date_dq_volume = fh.transform_large_df(dq_volume)
    date_shr_today = fh.transform_large_df(shr_today)

    turnover = date_dq_volume / date_shr_today
    turnover = turnover.loc[list(date_dq_volume.index)]
    # 在中国:换手率= 成交量/流通股本×100%,炒股软件的成交量单位是万,流通股本单位是亿
    # 流通股本基本不变(大概1年变一次),我都取得各自的总量,然后相除
    format = lambda x: '%.4f' % x
    turnover_1m = turnover[:m1].sum()
    turnover_2m = turnover[:m2].sum()
    turnover_3m = turnover[:m3].sum()
    turnover_6m = turnover[:m6].sum()
    turnover_12m = turnover[:m12].sum()

    turnover_1m = turnover_1m[~turnover_1m.isin([np.nan, np.inf, -np.inf])]
    turnover_2m = turnover_2m[~turnover_2m.isin([np.nan, np.inf, -np.inf])]
    turnover_3m = turnover_3m[~turnover_3m.isin([np.nan, np.inf, -np.inf])]
    turnover_6m = turnover_6m[~turnover_6m.isin([np.nan, np.inf, -np.inf])]
    turnover_12m = turnover_12m[~turnover_12m.isin([np.nan, np.inf, -np.inf])]

    turnover_1m = turnover_1m.dropna()
    turnover_2m = turnover_2m.dropna()
    turnover_3m = turnover_3m.dropna()
    turnover_6m = turnover_6m.dropna()
    turnover_12m = turnover_12m.dropna()

    turnover_1m = turnover_1m.map(format)
    turnover_2m = turnover_2m.map(format)
    turnover_3m = turnover_3m.map(format)
    turnover_6m = turnover_6m.map(format)
    turnover_12m = turnover_12m.map(format)

    # 入库
    stock_list = om.get_all_stock(cur_date)
    turnover_1m_flag = fh.df_to_list(cur_date, turnover_1m, 'turnover_1m', stock_list)
    turnover_2m_flag = fh.df_to_list(cur_date, turnover_2m, 'turnover_2m', stock_list)
    turnover_3m_flag = fh.df_to_list(cur_date, turnover_3m, 'turnover_3m', stock_list)
    turnover_6m_flag = fh.df_to_list(cur_date, turnover_6m, 'turnover_6m', stock_list)
    turnover_12m_flag = fh.df_to_list(cur_date, turnover_12m, 'turnover_12m', stock_list)

    if turnover_1m_flag & turnover_2m_flag & turnover_3m_flag & turnover_6m_flag & turnover_12m_flag != 1:
        gl.logger.error('factor_turnover insert fail')
        return 0
    else:
        gl.logger.info('factor_turnover insert successful')
        return 1
예제 #5
0
def factor_financial_quality(cur_date):
    # 获取经营性现金流(ttm)
    val_pcf_ocfttm = om.get_index_val_pcf_ocfttm(cur_date)
    # 单季度.净资产收益率,-1列为rank,-2列为所需要的
    qfa_roe = om.get_qfa_roe(cur_date)
    # 净资产收益率(ttm),-1列为rank,-2列为所需要的
    fa_roe = om.get_fa_roe(cur_date)
    # 单季度.总资产净利润,-1列为rank,-2列为所需要的
    qfa_roa = om.get_qfa_roa(cur_date)
    # 总资产净利润(ttm),-1列为rank,-2列为所需要的
    fa_roa = om.get_fa_roa(cur_date)
    # 单季度.销售毛利率,-1列为rank,-2列为所需要的
    qfa_grossprofitmargin = om.get_qfa_grossprofitmargin(cur_date)
    # 销售毛利率(ttm),-1列为rank,-2列为所需要的
    fa_grossprofitmargin = om.get_fa_grossprofitmargin(cur_date)
    # 扣除非经常损益后的净利润/净利润,-1列为rank,-2列为所需要的
    deductedprofittoprofit = om.get_deductedprofittoprofit(cur_date)
    # 总资产周转率(ttm),-1列为rank,-2列为所需要的
    fa_assetsturn = om.get_fa_assetsturn(cur_date)
    # 获取经营性现金流(lyr)
    val_pcf_ocf = om.get_index_val_pcf_ocf(cur_date)

    roe_q = qfa_roe.iloc[:, -2]
    roe_ttm = fa_roe.iloc[:, -2]
    roa_q = qfa_roa.iloc[:, -2]
    roa_ttm = fa_roa.iloc[:, -2]
    grossprofitmargin_q = qfa_grossprofitmargin.iloc[:, -2]
    grossprofitmargin_ttm = fa_grossprofitmargin.iloc[:, -2]
    profitmargin_ttm = deductedprofittoprofit.iloc[:, -2]
    assetturnover_ttm = fa_assetsturn.iloc[:, -2]
    operationcashflowratio_q = val_pcf_ocf.iloc[:, -1] / qfa_roa.iloc[:, -2]
    operationcashflowratio_ttm = val_pcf_ocfttm.iloc[:, -1] / fa_roa.iloc[:, -2]
    # 入库
    stock_list = om.get_all_stock(cur_date)
    roe_q_flag = fh.df_to_list(cur_date, roe_q, 'roe_q', stock_list)
    roe_ttm_flag = fh.df_to_list(cur_date, roe_ttm, 'roe_ttm', stock_list)
    roa_q_flag = fh.df_to_list(cur_date, roa_q, 'roa_q', stock_list)
    roa_ttm_flag = fh.df_to_list(cur_date, roa_ttm, 'roa_ttm', stock_list)
    grossprofitmargin_q_flag = fh.df_to_list(cur_date, grossprofitmargin_q, 'grossprofitmargin_q', stock_list)
    grossprofitmargin_ttm_flag = fh.df_to_list(cur_date, grossprofitmargin_ttm, 'grossprofitmargin_ttm', stock_list)
    profitmargin_ttm_flag = fh.df_to_list(cur_date, profitmargin_ttm, 'profitmargin_ttm', stock_list)
    assetturnover_ttm_flag = fh.df_to_list(cur_date, assetturnover_ttm, 'assetturnover_ttm', stock_list)
    operationcashflowratio_q_flag = fh.df_to_list(cur_date, operationcashflowratio_q, 'operationcashflowratio_q', stock_list)
    operationcashflowratio_ttm_flag = fh.df_to_list(cur_date, operationcashflowratio_ttm, 'operationcashflowratio_ttm', stock_list)

    if roe_q_flag & roe_ttm_flag & roa_q_flag & roa_ttm_flag & grossprofitmargin_q_flag & grossprofitmargin_ttm_flag \
            & profitmargin_ttm_flag & assetturnover_ttm_flag & operationcashflowratio_q_flag & operationcashflowratio_ttm_flag != 1:
        gl.logger.error('factor_financial_quality insert fail')
        return 0
    else:
        gl.logger.info('factor_financial_quality insert successful')
        return 1
예제 #6
0
def factor_volatility(cur_date, date_list, regress_result, date_adjclose):
    m1, m2, m3, m6, m12 = 20, 40, 60, 120, 240
    # 获取N天交易日中每天的最高价和最低价,不带index,为了格式化df成为index为日期,columns为股票代码
    price_high = om.get_ndays_dq_high(cur_date, date_list)
    price_low = om.get_ndays_dq_low(cur_date, date_list)
    # 得到近一年的最高价,最低价,并转换为index为日期,columns为股票代码
    date_price_high = fh.transform_large_df(price_high)
    date_price_low = fh.transform_large_df(price_low)
    # 降序排列index(最上面的是最新的日期), 结果为20天内最高价和最低价
    # 获取N天交易日中每天的最高价,不带index,为了格式化df成为index为日期,columns为股票代码
    price_max_1m, price_min_1m = price(date_price_high, date_price_low, m1)
    price_max_2m, price_min_2m = price(date_price_high, date_price_low, m2)
    price_max_3m, price_min_3m = price(date_price_high, date_price_low, m3)
    price_max_6m, price_min_6m = price(date_price_high, date_price_low, m6)
    price_max_12m, price_min_12m = price(date_price_high, date_price_low, m12)

    high_low_1m = price_max_1m / price_min_1m
    high_low_2m = price_max_2m / price_min_2m
    high_low_3m = price_max_3m / price_min_3m
    high_low_6m = price_max_6m / price_min_6m
    high_low_12m = price_max_12m / price_min_12m
    std_1m = std_adjclose_rate(date_adjclose, m1)
    std_2m = std_adjclose_rate(date_adjclose, m2)
    std_3m = std_adjclose_rate(date_adjclose, m3)
    std_6m = std_adjclose_rate(date_adjclose, m6)
    std_12m = std_adjclose_rate(date_adjclose, m12)
    ln_price = np.log(date_adjclose.iloc[0])  # 用的每日收盘价
    beta_consistence = regress_result.loc['beta_consis']
    # 入库
    stock_list = om.get_all_stock(cur_date)
    high_low_1m_flag = fh.df_to_list(cur_date, high_low_1m, 'high_low_1m', stock_list)
    high_low_2m_flag = fh.df_to_list(cur_date, high_low_2m, 'high_low_2m', stock_list)
    high_low_3m_flag = fh.df_to_list(cur_date, high_low_3m, 'high_low_3m', stock_list)
    high_low_6m_flag = fh.df_to_list(cur_date, high_low_6m, 'high_low_6m', stock_list)
    high_low_12m_flag = fh.df_to_list(cur_date, high_low_12m, 'high_low_12m', stock_list)
    std_1m_flag = fh.df_to_list(cur_date, std_1m, 'std_1m', stock_list)
    std_2m_flag = fh.df_to_list(cur_date, std_2m, 'std_2m', stock_list)
    std_3m_flag = fh.df_to_list(cur_date, std_3m, 'std_3m', stock_list)
    std_6m_flag = fh.df_to_list(cur_date, std_6m, 'std_6m', stock_list)
    std_12m_flag = fh.df_to_list(cur_date, std_12m, 'std_12m', stock_list)
    ln_price_flag = fh.df_to_list(cur_date, ln_price, 'ln_price', stock_list)
    beta_consistence_flag = fh.df_to_list(cur_date, beta_consistence, 'beta_consistence', stock_list)

    if high_low_1m_flag & high_low_2m_flag & high_low_3m_flag & high_low_6m_flag & high_low_12m_flag & \
            std_1m_flag & std_2m_flag & std_3m_flag & std_6m_flag & std_12m_flag & ln_price_flag & beta_consistence_flag != 1:
        gl.logger.error('factor_volatility insert fail')
        return 0
    else:
        gl.logger.info('factor_volatility insert successful')
        return 1
예제 #7
0
def factor_value(cur_date):
    # 获取EP(TTM),净利润/总市值(带index),(若净利润<=0,则返回空)
    inverse_ep = om.get_index_ep(cur_date)
    # 获取ps,市销率(PS,TTM)
    ps = om.get_ps(cur_date)
    # 获取pb值  也就是需要的bp,的倒数
    inverse_bp = om.get_index_val_pb_new(cur_date)
    # 获取净现金流
    val_pcf_ncfttm = om.get_index_val_pcf_ncfttm(cur_date)
    # 获取经营性现金流(ttm)
    val_pcf_ocfttm = om.get_index_val_pcf_ocfttm(cur_date)
    # 获取个股市值(带index)
    index_mkvalue = om.get_index_mkvalue(cur_date)

    ep = 1 / inverse_ep.iloc[:, -1]
    bp = 1 / inverse_bp.iloc[:, -1]
    sp = 1 / ps.iloc[:, -1]
    ncfp = val_pcf_ncfttm.iloc[:, -1] / index_mkvalue.iloc[:, -1]
    ocfp = val_pcf_ocfttm.iloc[:, -1] / index_mkvalue.iloc[:, -1]
    # 分红放在最后面了,主要是不是我写的
    # 红利(近12个月股息率),首先拼接一年前的日期
    struct_time = time.strptime(str(cur_date), '%Y%m%d')
    y = struct_time.tm_year - 1
    m = struct_time.tm_mon
    d = struct_time.tm_mday
    year_before = str(date(y, m, d).replace(year=y))
    year_before = year_before.replace('-', '')
    divdend = om.get_dps(year_before, str(cur_date))
    divdend.index = divdend.iloc[:, 0]
    dp = divdend.iloc[:, -1]
    # 入库
    stock_list = om.get_all_stock(cur_date)
    ep_flag = fh.df_to_list(cur_date, ep, 'ep', stock_list)
    bp_flag = fh.df_to_list(cur_date, bp, 'bp', stock_list)
    sp_flag = fh.df_to_list(cur_date, sp, 'sp', stock_list)
    ncfp_flag = fh.df_to_list(cur_date, ncfp, 'ncfp', stock_list)
    ocfp_flag = fh.df_to_list(cur_date, ocfp, 'ocfp', stock_list)
    dp_flag = fh.df_to_list(cur_date, dp, 'dp', stock_list)

    if ep_flag & bp_flag & sp_flag & ncfp_flag & ocfp_flag & dp_flag != 1:
        gl.logger.error('factor_value insert fail')
        return 0
    else:
        gl.logger.info('factor_value insert successful')
        return 1
예제 #8
0
def factor_modified_momentum(cur_date, date_list, date_adjclose, date_dq_turn):
    m1, m2, m3, m6, m12 = 20, 40, 60, 120, 240
    weighted_strength_1m = calculate_rate(date_adjclose, date_dq_turn, m1)
    weighted_strength_2m = calculate_rate(date_adjclose, date_dq_turn, m2)
    weighted_strength_3m = calculate_rate(date_adjclose, date_dq_turn, m3)
    weighted_strength_6m = calculate_rate(date_adjclose, date_dq_turn, m6)
    weighted_strength_12m = calculate_rate(date_adjclose, date_dq_turn, m12)
    # 入库
    stock_list = om.get_all_stock(cur_date)
    weighted_strength_1m_flag = fh.df_to_list(cur_date, weighted_strength_1m, 'weighted_strength_1m', stock_list)
    weighted_strength_2m_flag = fh.df_to_list(cur_date, weighted_strength_2m, 'weighted_strength_2m', stock_list)
    weighted_strength_3m_flag = fh.df_to_list(cur_date, weighted_strength_3m, 'weighted_strength_3m', stock_list)
    weighted_strength_6m_flag = fh.df_to_list(cur_date, weighted_strength_6m, 'weighted_strength_6m', stock_list)
    weighted_strength_12m_flag = fh.df_to_list(cur_date, weighted_strength_12m, 'weighted_strength_12m', stock_list)

    if weighted_strength_1m_flag & weighted_strength_2m_flag & weighted_strength_3m_flag & weighted_strength_6m_flag & \
            weighted_strength_12m_flag != 1:
        gl.logger.error('factor_modified_momentum insert fail')
        return 0
    else:
        gl.logger.info('factor_modified_momentum insert successful')
        return 1
예제 #9
0
def factor_momentum(cur_date, regress_result, date_adjclose):
    HAlpha = regress_result.loc['const']
    relative_strength_1m = date_adjclose.iloc[0] / date_adjclose.shift(-20).iloc[0] - 1
    relative_strength_2m = date_adjclose.iloc[0] / date_adjclose.shift(-20 * 2).iloc[0] - 1
    relative_strength_3m = date_adjclose.iloc[0] / date_adjclose.shift(-20 * 3).iloc[0] - 1
    relative_strength_6m = date_adjclose.iloc[0] / date_adjclose.shift(-20 * 6).iloc[0] - 1
    relative_strength_12m = date_adjclose.iloc[0] / date_adjclose.shift(-20 * 12).iloc[0] - 1
    # 入库
    stock_list = om.get_all_stock(cur_date)
    HAlpha_flag = fh.df_to_list(cur_date, HAlpha, 'HAlpha', stock_list)
    relative_strength_1m_flag = fh.df_to_list(cur_date, relative_strength_1m, 'relative_strength_1m', stock_list)
    relative_strength_2m_flag = fh.df_to_list(cur_date, relative_strength_2m, 'relative_strength_2m', stock_list)
    relative_strength_3m_flag = fh.df_to_list(cur_date, relative_strength_3m, 'relative_strength_3m', stock_list)
    relative_strength_6m_flag = fh.df_to_list(cur_date, relative_strength_6m, 'relative_strength_6m', stock_list)
    relative_strength_12m_flag = fh.df_to_list(cur_date, relative_strength_12m, 'relative_strength_12m', stock_list)

    if HAlpha_flag & relative_strength_1m_flag & relative_strength_2m_flag & relative_strength_3m_flag & \
            relative_strength_6m_flag & relative_strength_12m_flag != 1:
        gl.logger.error('factor_momentum insert fail')
        return 0
    else:
        gl.logger.info('factor_momentum insert successful')
        return 1