Exemplo n.º 1
0
    def bond_var(period, percent, day):
        
        df_nav = base_ra_index_nav.load_series('120000011')
        df_inc = df_nav.pct_change(period)
        df_inc = df_inc.dropna()
        df_inc = df_inc[df_inc.index < day]
        df_inc = df_inc.iloc[-365*5:]
        var = np.percentile(df_inc, percent)

        #ib_inc = Asset('PO.IB0010').nav().pct_change()
        #lrb_inc = Asset('PO.LRB010').nav().pct_change()
        #dates = ib_inc.index & lrb_inc.index
        #ib_inc = ib_inc.loc[dates]
        #lrb_inc = lrb_inc.loc[dates]
        #ib_lrb_inc = ib_inc * 0.2 + lrb_inc * 0.8
        #ib_lrb_inc = ib_lrb_inc.reindex(df_inc.index).fillna(0.0)
        #ib_lrb_inc[ib_lrb_inc == 0] = df_inc[ib_lrb_inc == 0]
        #df_inc = ib_lrb_inc
        #df_nav = (df_inc + 1).cumprod()
        #df_inc = df_nav.pct_change(period).fillna(0.0)
        #df_inc = df_inc[df_inc.index < day]
        #df_inc = df_inc.iloc[-365*5:]
        #var = np.percentile(df_inc, percent)

        return var
Exemplo n.º 2
0
def macroview_retcompare(ctx, st_date, ed_date):
    startDate = parse(st_date)
    endDate = parse(ed_date)
    assetsID = {
        '120000001': '沪深300',
        '120000002': '中证500',
        '120000013': '标普500',
        '120000015': '恒生指数',
        '120000014': '沪金指数',
        '120000010': '中证国债',
        '120000011': '中证信用债'
    }
    #assetsRet = {'120000001':0.2743,'120000002':0.3576,'120000013':0.0953,'120000015':0.0997,'120000014':-0.0029,'120000010':0.0009,'120000011':0.0032}
    assets = dict([(asset_id, base_ra_index_nav.load_series(asset_id))
                   for asset_id in list(assetsID.keys())])
    df_assets = pd.DataFrame(assets).loc[startDate:endDate, ].fillna(
        method='pad')
    assetsRet = dict([(
        asset_id,
        df_assets.loc[endDate, asset_id] / df_assets.loc[startDate, asset_id] -
        1.0) for asset_id in list(assetsID.keys())])
    #     #计算给定指数和日期的持有期收益
    df_assets = pd.DataFrame(assets)
    df_assets = df_assets.rolling(365).apply(lambda x: x[-1] / x[0] - 1,
                                             raw=True)
    df = df_assets.reset_index()
    tradedaysDiff = df[df.date == endDate].index.tolist()[0] - df[
        df.date == startDate].index.tolist()[0]
    annualMulty = 365 / tradedaysDiff
    MacroCompare = []
    for key, value in assetsRet.items():
        values = value * annualMulty
        ser = df_assets[key].dropna()
        ser = ser.sort_values()
        if len(ser) == 0:
            continue
        MacroCompare.append(
            (assetsID[key], 1 - len(ser[ser < values]) / len(ser)))
    df_MacroCompare = pd.DataFrame(MacroCompare, columns=['ra_index', '分位数'])
    #print(df_MacroCompare)
    ser = pd.Series(assetsRet)
    ret_df = pd.DataFrame(columns=ser.index)
    ret_df.loc[endDate] = ser
    ret_df = ret_df.rename(columns=assetsID)
    print(ret_df)
    df_MacroCompare['date'] = endDate
    df_MacroCompare = df_MacroCompare.set_index(['date', 'ra_index']).unstack()
    df_MacroCompare.columns = df_MacroCompare.columns.get_level_values(1)
    df_MacroCompare = df_MacroCompare[ret_df.columns]
    df_MacroCompare.columns = df_MacroCompare.columns + ' 分位数'
    df = pd.concat([ret_df, df_MacroCompare], axis=1)
    print(df)
    df.to_csv('宏观观点数据.csv', encoding='gbk')
Exemplo n.º 3
0
    def load_nav_series(asset_id,
                        reindex=None,
                        begin_date=None,
                        end_date=None):

        prefix = asset_id[0:2]
        if prefix.isdigit():
            xtype = int(asset_id) // 10000000
            if xtype == 1:
                #
                # 基金池资产
                #
                asset_id = int(asset_id) % 10000000
                (pool_id, category) = (asset_id // 100, asset_id % 100)
                ttype = pool_id // 10000
                sr = asset_ra_pool_nav.load_series(pool_id,
                                                   category,
                                                   ttype,
                                                   reindex=reindex,
                                                   begin_date=begin_date,
                                                   end_date=end_date)
            elif xtype == 3:
                #
                # 基金池资产
                #
                sr = base_ra_fund_nav.load_series(asset_id,
                                                  reindex=reindex,
                                                  begin_date=begin_date,
                                                  end_date=end_date)
            elif xtype == 4:
                #
                # 修型资产
                #
                sr = asset_rs_reshape_nav.load_series(asset_id,
                                                      reindex=reindex,
                                                      begin_date=begin_date,
                                                      end_date=end_date)
            elif xtype == 12:
                #
                # 指数资产
                #
                sr = base_ra_index_nav.load_series(asset_id,
                                                   reindex=reindex,
                                                   begin_date=begin_date,
                                                   end_date=end_date)
            else:
                sr = pd.Series()
        else:
            if prefix == 'AP':
                #
                # 基金池资产
                #
                sr = asset_ra_pool_nav.load_series(asset_id,
                                                   0,
                                                   9,
                                                   reindex=reindex,
                                                   begin_date=begin_date,
                                                   end_date=end_date)
            elif prefix == 'FD':
                #
                # 基金资产
                #
                sr = base_ra_fund_nav.load_series(asset_id,
                                                  reindex=reindex,
                                                  begin_date=begin_date,
                                                  end_date=end_date)
            elif prefix == 'RS':
                #
                # 修型资产
                #
                sr = asset_rs_reshape_nav.load_series(asset_id,
                                                      reindex=reindex,
                                                      begin_date=begin_date,
                                                      end_date=end_date)
            elif prefix == 'IX':
                #
                # 指数资产
                #
                sr = base_ra_index_nav.load_series(asset_id,
                                                   reindex=reindex,
                                                   begin_date=begin_date,
                                                   end_date=end_date)
            elif prefix == 'ER':
                sr = base_exchange_rate_index_nav.load_series(
                    asset_id,
                    reindex=reindex,
                    begin_date=begin_date,
                    end_date=end_date)
            elif prefix == 'SK':
                #
                # 股票资产
                #
                sr = db.asset_stock.load_stock_nav_series(
                    asset_id,
                    reindex=reindex,
                    begin_date=begin_date,
                    end_date=end_date)
            elif prefix == 'MZ':
                #
                # Markowitz asset
                #
                sr = db.asset_mz_markowitz_nav.load_series(
                    asset_id,
                    reindex=reindex,
                    begin_date=begin_date,
                    end_date=end_date)
            elif prefix == 'PO':
                #
                # Markowitz asset
                #
                sr = db.asset_ra_portfolio_nav.load_series(
                    asset_id,
                    9,
                    reindex=reindex,
                    begin_date=begin_date,
                    end_date=end_date)
            else:
                sr = pd.Series()

        return sr
Exemplo n.º 4
0
def allocate_benchmark_comp(ctx):
    index_ids = ['120000016', '120000010']
    data = {}
    for _id in index_ids:
        data[_id] = base_ra_index_nav.load_series(_id)
    df = pd.DataFrame(data)

    composite_asset_ids = [
        '20201', '20202', '20203', '20204', '20205', '20206', '20207', '20208'
    ]

    data = {}

    for _id in composite_asset_ids:
        nav = asset_ra_composite_asset_nav.load_nav(_id)
        nav = nav.reset_index()
        nav = nav[['ra_date', 'ra_nav']]
        nav = nav.set_index(['ra_date'])
        data[_id] = nav.ra_nav

    bench_df = pd.DataFrame(data)
    benchmark_df = pd.concat([bench_df, df],
                             axis=1,
                             join_axes=[bench_df.index])

    conn = MySQLdb.connect(**config.db_asset)
    conn.autocommit(True)

    dfs = []
    for i in range(0, 10):
        sql = 'select on_date as date, on_nav as nav from on_online_nav where on_online_id = 80000%d and on_type = 8' % i
        df = pd.read_sql(sql, conn, index_col=['date'], parse_dates=['date'])
        df.columns = ['risk_' + str(i)]
        dfs.append(df)

    df = pd.concat(dfs, axis=1)

    conn.close()

    df = pd.concat([df, benchmark_df], axis=1, join_axes=[df.index])
    df = df.fillna(method='pad')
    df = df.rename(
        columns={
            'risk_0': '风险10',
            'risk_1': '风险1',
            'risk_2': '风险2',
            'risk_3': '风险3',
            'risk_4': '风险4',
            'risk_5': '风险5',
            'risk_6': '风险6',
            'risk_7': '风险7',
            'risk_8': '风险8',
            'risk_9': '风险9',
            '20201': '风险2比较基准',
            '20202': '风险3比较基准',
            '20203': '风险4比较基准',
            '20204': '风险5比较基准',
            '20205': '风险6比较基准',
            '20206': '风险7比较基准',
            '20207': '风险8比较基准',
            '20208': '风险9比较基准',
            '120000016': '风险10比较基准',
            '120000010': '风险1比较基准'
        })
    cols = [
        '风险1', '风险2', '风险3', '风险4', '风险5', '风险6', '风险7', '风险8', '风险9', '风险10',
        '风险1比较基准', '风险2比较基准', '风险3比较基准', '风险4比较基准', '风险5比较基准', '风险6比较基准',
        '风险7比较基准', '风险8比较基准', '风险9比较基准', '风险10比较基准'
    ]
    df = df[cols]

    #df.to_csv('online_benchmark.csv', encoding='gbk')

    result_df = pd.DataFrame(columns=df.columns)
    last_day = df.index[-1]
    #print(last_day)
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 当日'] = df.pct_change().iloc[-1]
    result_df.loc[
        df.index[-1].strftime('%Y-%m-%d') +
        ' 过去一周'] = df.loc[last_day] / df.loc[last_day - timedelta(weeks=1)] - 1
    result_df.loc[
        df.index[-1].strftime('%Y-%m-%d') +
        ' 过去一月'] = df.loc[last_day] / df.loc[last_day - timedelta(days=31)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去三个月'] = df.loc[last_day] / df.loc[last_day -
                                                        timedelta(days=91)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去六个月'] = df.loc[last_day] / df.loc[last_day - timedelta(
                      days=182)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去一年'] = df.loc[last_day] / df.loc[last_day -
                                                       timedelta(days=365)] - 1
    #result_df.to_csv('智能组合收益与比较基准收益比较.csv', encoding='gbk')

    df = result_df

    df = df.reset_index()
    df['date'] = pd.Series(None)
    df['statistics_type'] = pd.Series(None)
    df = df.apply(split_ct, axis=1)
    df.pop('index')
    df.set_index('date', inplace=True)
    cols_list = list(df.columns)
    cols_list.insert(0, cols_list.pop(-1))
    df = df[cols_list]

    if len(df) < 1:
        print("无数据需要插入")
        return
    # 开始注入数据
    engine = database.connection('asset')
    Session = sessionmaker(bind=engine)
    session = Session()

    for i in range(len(df)):
        ins = asset_allocate.smart_risk(sir_date=df.index[i],
                                        sir_statistics_type=df.iloc[i, 0],
                                        sir_risk1=df.iloc[i, 1],
                                        sir_risk2=df.iloc[i, 2],
                                        sir_risk3=df.iloc[i, 3],
                                        sir_risk4=df.iloc[i, 4],
                                        sir_risk5=df.iloc[i, 5],
                                        sir_risk6=df.iloc[i, 6],
                                        sir_risk7=df.iloc[i, 7],
                                        sir_risk8=df.iloc[i, 8],
                                        sir_risk9=df.iloc[i, 9],
                                        sir_risk10=df.iloc[i, 10],
                                        sir_risk1_standard=df.iloc[i, 11],
                                        sir_risk2_standard=df.iloc[i, 12],
                                        sir_risk3_standard=df.iloc[i, 13],
                                        sir_risk4_standard=df.iloc[i, 14],
                                        sir_risk5_standard=df.iloc[i, 15],
                                        sir_risk6_standard=df.iloc[i, 16],
                                        sir_risk7_standard=df.iloc[i, 17],
                                        sir_risk8_standard=df.iloc[i, 18],
                                        sir_risk9_standard=df.iloc[i, 19],
                                        sir_risk10_standard=df.iloc[i, 20])
        session.add(ins)
        session.commit()
        print("成功插入了 %d 条数据;" % (i + 1))
Exemplo n.º 5
0
def macroview_retcompare(ctx, st_date, ed_date):
    startDate = parse(st_date)
    endDate = parse(ed_date)
    assetsID = {
        '120000001': '沪深300',
        '120000002': '中证500',
        '120000013': '标普500',
        '120000015': '恒生指数',
        '120000014': '沪金指数',
        '120000010': '中证国债',
        '120000011': '中证信用债'
    }
    #assetsRet = {'120000001':0.2743,'120000002':0.3576,'120000013':0.0953,'120000015':0.0997,'120000014':-0.0029,'120000010':0.0009,'120000011':0.0032}
    assets = dict([(asset_id, base_ra_index_nav.load_series(asset_id))
                   for asset_id in list(assetsID.keys())])
    df_assets = pd.DataFrame(assets).loc[startDate:endDate, ].fillna(
        method='pad')
    assetsRet = dict([(
        asset_id,
        df_assets.loc[endDate, asset_id] / df_assets.loc[startDate, asset_id] -
        1.0) for asset_id in list(assetsID.keys())])
    #     #计算给定指数和日期的持有期收益
    df_assets = pd.DataFrame(assets)
    df_assets = df_assets.rolling(365).apply(lambda x: x[-1] / x[0] - 1,
                                             raw=True)
    df = df_assets.reset_index()
    tradedaysDiff = df[df.date == endDate].index.tolist()[0] - df[
        df.date == startDate].index.tolist()[0]
    annualMulty = 365 / tradedaysDiff
    MacroCompare = []
    for key, value in assetsRet.items():
        values = value * annualMulty
        ser = df_assets[key].dropna()
        ser = ser.sort_values()
        if len(ser) == 0:
            continue
        MacroCompare.append(
            (assetsID[key], 1 - len(ser[ser < values]) / len(ser)))
    df_MacroCompare = pd.DataFrame(MacroCompare, columns=['ra_index', '分位数'])
    #print(df_MacroCompare)
    ser = pd.Series(assetsRet)
    ret_df = pd.DataFrame(columns=ser.index)
    ret_df.loc[endDate] = ser
    ret_df = ret_df.rename(columns=assetsID)
    print(ret_df)
    df_MacroCompare['date'] = endDate
    df_MacroCompare = df_MacroCompare.set_index(['date', 'ra_index']).unstack()
    df_MacroCompare.columns = df_MacroCompare.columns.get_level_values(1)
    df_MacroCompare = df_MacroCompare[ret_df.columns]
    df_MacroCompare.columns = df_MacroCompare.columns + ' 分位数'
    df = pd.concat([ret_df, df_MacroCompare], axis=1)
    print(df)

    # 开始注入数据
    if len(df) < 1:
        print("无数据需要插入")
        return

    engine = database.connection('asset')
    Session = sessionmaker(bind=engine)
    session = Session()

    for count, index in enumerate(df.index):
        ins = asset_allocate.opinion(
            #id = None,
            oi_date=index,
            oi_return_CSI300=df.loc[index, df.columns[0]],
            oi_return_IC500=df.loc[index, df.columns[1]],
            oi_return_SP500=df.loc[index, df.columns[2]],
            oi_return_HSI=df.loc[index, df.columns[3]],
            oi_return_SQau=df.loc[index, df.columns[4]],
            oi_return_CSI_TB=df.loc[index, df.columns[5]],
            oi_return_CSI_DB=df.loc[index, df.columns[6]],
            #oi_comment = None
        )
        session.add(ins)
        session.commit()
        print("成功插入 %d 条数据;" % (count + 1))

    session.close()
Exemplo n.º 6
0
def allocate_benchmark_comp(ctx):

    index_ids = ['120000016', '120000010']
    data = {}
    for _id in index_ids:
        data[_id] = base_ra_index_nav.load_series(_id)
    df = pd.DataFrame(data)

    composite_asset_ids = [
        '20201', '20202', '20203', '20204', '20205', '20206', '20207', '20208'
    ]

    data = {}

    for _id in composite_asset_ids:
        nav = asset_ra_composite_asset_nav.load_nav(_id)
        nav = nav.reset_index()
        nav = nav[['ra_date', 'ra_nav']]
        nav = nav.set_index(['ra_date'])
        data[_id] = nav.ra_nav

    bench_df = pd.DataFrame(data)
    benchmark_df = pd.concat([bench_df, df],
                             axis=1,
                             join_axes=[bench_df.index])

    conn = MySQLdb.connect(**config.db_asset)
    conn.autocommit(True)

    dfs = []
    for i in range(0, 10):
        sql = 'select on_date as date, on_nav as nav from on_online_nav where on_online_id = 80000%d and on_type = 8' % i
        df = pd.read_sql(sql, conn, index_col=['date'], parse_dates=['date'])
        df.columns = ['risk_' + str(i)]
        dfs.append(df)

    df = pd.concat(dfs, axis=1)

    conn.close()

    df = pd.concat([df, benchmark_df], axis=1, join_axes=[df.index])
    df = df.fillna(method='pad')
    df = df.rename(
        columns={
            'risk_0': '风险10',
            'risk_1': '风险1',
            'risk_2': '风险2',
            'risk_3': '风险3',
            'risk_4': '风险4',
            'risk_5': '风险5',
            'risk_6': '风险6',
            'risk_7': '风险7',
            'risk_8': '风险8',
            'risk_9': '风险9',
            '20201': '风险2比较基准',
            '20202': '风险3比较基准',
            '20203': '风险4比较基准',
            '20204': '风险5比较基准',
            '20205': '风险6比较基准',
            '20206': '风险7比较基准',
            '20207': '风险8比较基准',
            '20208': '风险9比较基准',
            '120000016': '风险10比较基准',
            '120000010': '风险1比较基准'
        })
    cols = [
        '风险1', '风险2', '风险3', '风险4', '风险5', '风险6', '风险7', '风险8', '风险9', '风险10',
        '风险1比较基准', '风险2比较基准', '风险3比较基准', '风险4比较基准', '风险5比较基准', '风险6比较基准',
        '风险7比较基准', '风险8比较基准', '风险9比较基准', '风险10比较基准'
    ]
    df = df[cols]

    result_df = pd.DataFrame(columns=df.columns)
    last_day = df.index[-1]
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 当日'] = df.pct_change().iloc[-1]
    result_df.loc[
        df.index[-1].strftime('%Y-%m-%d') +
        ' 过去一周'] = df.loc[last_day] / df.loc[last_day - timedelta(weeks=1)] - 1
    result_df.loc[
        df.index[-1].strftime('%Y-%m-%d') +
        ' 过去一月'] = df.loc[last_day] / df.loc[last_day - timedelta(days=31)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去三个月'] = df.loc[last_day] / df.loc[last_day -
                                                        timedelta(days=91)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去六个月'] = df.loc[last_day] / df.loc[last_day - timedelta(
                      days=182)] - 1
    result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                  ' 过去一年'] = df.loc[last_day] / df.loc[last_day -
                                                       timedelta(days=365)] - 1
    result_df.to_csv('智能组合收益与比较基准收益比较.csv', encoding='gbk')
Exemplo n.º 7
0
def cal_layer_type():
    engine = database.connection('asset')
    Session = sessionmaker(bind=engine)
    session = Session()

    sql1 = session.query(factor_cluster_asset.fc_asset_id).filter(
        factor_cluster_asset.globalid == 'FC.000004')
    asset_ids = [asset[0] for asset in sql1.all()]
    assets = {}
    for asset_id in asset_ids:
        assets[asset_id] = load_nav_series(asset_id)

    trade_dates = DBData.trade_dates(start_date='2010-06-01',
                                     end_date='2018-12-01')
    df_assets = pd.DataFrame(assets)
    # df_assets = df_assets[df_assets.index >= start_date]
    # df_assets = df_assets[df_assets.index <= end_date]
    df_assets = df_assets.reindex(trade_dates).dropna()
    df_ret = df_assets.pct_change().dropna()

    layer_info = pd.read_csv('data/df_result.csv',
                             index_col=0,
                             parse_dates=True)
    dates = layer_info.index.unique()

    sh300 = base_ra_index_nav.load_series('120000001')
    sh300 = sh300.reindex(trade_dates).pct_change().dropna()
    zz500 = base_ra_index_nav.load_series('120000002')
    zz500 = zz500.reindex(trade_dates).pct_change().dropna()
    cyb = base_ra_index_nav.load_series('120000018')
    cyb = cyb.reindex(trade_dates).pct_change().dropna()

    for i in range(12, len(dates)):
        start_date = dates[i - 12]
        end_date = dates[i]
        layers = layer_info.loc[end_date].layer.unique()
        tmp_sh300 = sh300.loc[start_date.date():end_date.date()]
        tmp_zz500 = zz500.loc[start_date.date():end_date.date()]
        tmp_cyb = cyb.loc[start_date.date():end_date.date()]
        for layer in layers:
            layer_factor_id = layer_info[layer_info.layer ==
                                         layer].loc[end_date].factor_id
            if type(layer_factor_id) == str:
                layer_factor_id = np.array([layer_factor_id])
            else:
                layer_factor_id = layer_factor_id.values

            layer_ret = df_ret.loc[start_date.date():end_date.date(),
                                   layer_factor_id]
            layer_ret = layer_ret.mean(1)
            # layer_nav = (1+layer_ret.mean(1)).cumprod()
            sh300_corr = np.corrcoef(layer_ret, tmp_sh300)[0][1]
            zz500_corr = np.corrcoef(layer_ret, tmp_zz500)[0][1]
            cyb_corr = np.corrcoef(layer_ret, tmp_cyb)[0][1]
            corr_list = [sh300_corr, zz500_corr, cyb_corr]
            max_corr = max(corr_list)
            if max_corr < 0.85:
                sign = 'unknown'
            elif sh300_corr == max_corr:
                sign = 'sh300'
            elif zz500_corr == max_corr:
                sign = 'zz500'
            elif cyb_corr == max_corr:
                sign = 'cybzs'

            print end_date, layer, sign, sh300_corr, zz500_corr, cyb_corr
Exemplo n.º 8
0
def allocate_benchmark_comp():

    index_ids = ['120000016', '120000010']
    data = {}
    for _id in index_ids:
        data[_id] = base_ra_index_nav.load_series(_id)
    df = pd.DataFrame(data)

    composite_asset_ids = [
        '20201', '20202', '20203', '20204', '20205', '20206', '20207', '20208'
    ]

    data = {}

    for _id in composite_asset_ids:
        nav = asset_ra_composite_asset_nav.load_nav(_id)
        nav = nav.reset_index()
        nav = nav[['ra_date', 'ra_nav']]
        nav = nav.set_index(['ra_date'])
        data[_id] = nav.ra_nav

    bench_df = pd.DataFrame(data)
    benchmark_df = pd.concat([bench_df, df],
                             axis=1,
                             join_axes=[bench_df.index])

    conn = MySQLdb.connect(**config.db_asset)
    conn.autocommit(True)

    dfs = []
    for i in range(0, 10):
        sql = 'select on_date as date, on_nav as nav from on_online_nav where on_online_id = 80000%d and on_type = 8' % i
        df = pd.read_sql(sql, conn, index_col=['date'], parse_dates=['date'])
        df.columns = ['risk_' + str(i)]
        dfs.append(df)

    df = pd.concat(dfs, axis=1)

    conn.close()

    df = pd.concat([df, benchmark_df], axis=1, join_axes=[df.index])
    df = df.fillna(method='pad')
    df = df.rename(
        columns={
            'risk_0': '风险10',
            'risk_1': '风险1',
            'risk_2': '风险2',
            'risk_3': '风险3',
            'risk_4': '风险4',
            'risk_5': '风险5',
            'risk_6': '风险6',
            'risk_7': '风险7',
            'risk_8': '风险8',
            'risk_9': '风险9',
            '20201': '风险2比较基准',
            '20202': '风险3比较基准',
            '20203': '风险4比较基准',
            '20204': '风险5比较基准',
            '20205': '风险6比较基准',
            '20206': '风险7比较基准',
            '20207': '风险8比较基准',
            '20208': '风险9比较基准',
            '120000016': '风险10比较基准',
            '120000010': '风险1比较基准'
        })
    cols = [
        '风险1', '风险2', '风险3', '风险4', '风险5', '风险6', '风险7', '风险8', '风险9', '风险10',
        '风险1比较基准', '风险2比较基准', '风险3比较基准', '风险4比较基准', '风险5比较基准', '风险6比较基准',
        '风险7比较基准', '风险8比较基准', '风险9比较基准', '风险10比较基准'
    ]
    df = df[cols]

    result_df = pd.DataFrame(columns=df.columns)
    last_day = df.index[-1]
    print(last_day)
    first_day = df.index[0]
    print(first_day)
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 当日'] = df.pct_change().iloc[-1]
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一周'] = df.loc[last_day] / df.loc[last_day - timedelta(weeks = 1)] - 1
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 31)] - 1
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去三个月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 91)] - 1
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去六个月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 182)] - 1
    #result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一年'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 365)] - 1

    # 非滚动战胜概率
    length = len(df) // 7
    #    length = len(df)//31
    #    length = len(df)//91
    #    length = len(df)//182
    #    length = len(df)//365
    k1 = 0
    k2 = 0
    k3 = 0
    k4 = 0
    k5 = 0
    k6 = 0
    k7 = 0
    k8 = 0
    k9 = 0
    k10 = 0
    days = df.index.values
    for i in range(length):
        num = -1 - i * 7
        #        num = -1 - i*31
        #        num = -1 - i*91
        #        num = -1 - i*182
        #        num = -1 - i*365
        last_day = pd.to_datetime(str(days[num]))
        result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                      ' 过去一周'] = df.loc[last_day] / df.loc[
                          last_day - timedelta(weeks=1)] - 1
        #       result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 31)] - 1
        #       result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去三个月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 91)] - 1
        #       result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去六个月'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 182)] - 1
        #       result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一年'] = df.loc[last_day] / df.loc[last_day - timedelta(days = 365)] - 1
        if result_df['风险1'][0] > result_df['风险1比较基准'][0]:
            k1 += 1
        if result_df['风险2'][0] > result_df['风险2比较基准'][0]:
            k2 += 1
        if result_df['风险3'][0] > result_df['风险3比较基准'][0]:
            k3 += 1
        if result_df['风险4'][0] > result_df['风险4比较基准'][0]:
            k4 += 1
        if result_df['风险5'][0] > result_df['风险5比较基准'][0]:
            k5 += 1
        if result_df['风险6'][0] > result_df['风险6比较基准'][0]:
            k6 += 1
        if result_df['风险7'][0] > result_df['风险7比较基准'][0]:
            k7 += 1
        if result_df['风险8'][0] > result_df['风险8比较基准'][0]:
            k8 += 1
        if result_df['风险9'][0] > result_df['风险9比较基准'][0]:
            k9 += 1
        if result_df['风险10'][0] > result_df['风险10比较基准'][0]:
            k10 += 1


# 滚动战胜概率
#    length = len(df)-7
#    length = len(df)//7
    length = len(df) // 30
    k1 = 0
    k2 = 0
    k3 = 0
    k4 = 0
    k5 = 0
    k6 = 0
    k7 = 0
    k8 = 0
    k9 = 0
    k10 = 0
    days = df.index.values
    for i in range(length):
        #        num = -1 - i
        #        num = -1 - i*7
        #        num = -1 - i*30
        last_day = pd.to_datetime(str(days[num]))
        delta_day = last_day - timedelta(days=31)
        choose_dates = df[df.index <= delta_day].index.values
        if len(choose_dates) == 0:
            break
        delta_day = pd.to_datetime(str(choose_dates.max()))
        #        result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去一周'] = df.loc[last_day] / df.loc[delta_day] - 1
        result_df.loc[df.index[-1].strftime('%Y-%m-%d') +
                      ' 过去一月'] = df.loc[last_day] / df.loc[delta_day] - 1
        #        result_df.loc[df.index[-1].strftime('%Y-%m-%d') + ' 过去三个月'] = df.loc[last_day] / df.loc[delta_day] - 1
        if result_df['风险1'][0] > result_df['风险1比较基准'][0]:
            k1 += 1
        if result_df['风险2'][0] > result_df['风险2比较基准'][0]:
            k2 += 1
        if result_df['风险3'][0] > result_df['风险3比较基准'][0]:
            k3 += 1
        if result_df['风险4'][0] > result_df['风险4比较基准'][0]:
            k4 += 1
        if result_df['风险5'][0] > result_df['风险5比较基准'][0]:
            k5 += 1
        if result_df['风险6'][0] > result_df['风险6比较基准'][0]:
            k6 += 1
        if result_df['风险7'][0] > result_df['风险7比较基准'][0]:
            k7 += 1
        if result_df['风险8'][0] > result_df['风险8比较基准'][0]:
            k8 += 1
        if result_df['风险9'][0] > result_df['风险9比较基准'][0]:
            k9 += 1
        if result_df['风险10'][0] > result_df['风险10比较基准'][0]:
            k10 += 1

    r1 = k1 / length
    r2 = k2 / length
    r3 = k3 / length
    r4 = k4 / length
    r5 = k5 / length
    r6 = k6 / length
    r7 = k7 / length
    r8 = k8 / length
    r9 = k9 / length
    r10 = k10 / length

    print([r1, r2, r3, r4, r5, r6, r7, r8, r9, r10])