Add files via upload

Author: HaoningChen

scutquant/account.py

@@ -25,7 +25,7 @@ class Account:
 
     def __init__(self, init_cash: float, position: dict, available: dict, init_price: dict):
         self.cash = init_cash  # 可用资金
-        self.cash_available = deepcopy(init_cash)
+        # self.cash_available = deepcopy(init_cash)
         self.position = position  # keys应包括所有资产，如无头寸则值为0，以便按照keys更新持仓
         self.available = available  # 需要持有投资组合的底仓，否则按照T+1制度无法做空
         self.price = init_price  # 资产价格
@@ -39,7 +39,7 @@ def __init__(self, init_cash: float, position: dict, available: dict, init_price
         self.turnover = []
         self.trade_value = 0.0
 
-    def generate_total_order(self, order: dict, freq: int) -> dict:
+    def adjust_order(self, order: dict, freq: int) -> dict:
         order_offset = self.auto_offset(freq)
         if order_offset is not None:
             for key in order_offset["buy"].keys():
@@ -55,8 +55,8 @@ def generate_total_order(self, order: dict, freq: int) -> dict:
                     order["sell"][key] += order_offset["sell"][key]
         return order
 
-    def check_order(self, order: dict, price: dict, cost_rate: float = 0.0015, min_cost: float = 5) -> \
-            tuple[dict, bool]:  # 检查是否有足够的资金完成order, 如果不够则不买
+    def check_order(self, order: dict, price: dict, cost_rate: float = 0.0015, min_cost: float = 5) -> dict:
+        # 检查是否有足够的资金完成order, 如果不够则调整订单(sell不变, buy按比例减少)
         cash_inflow = 0.0
         cash_outflow = 0.0
         order_copy = deepcopy(order)
@@ -74,11 +74,13 @@ def check_order(self, order: dict, price: dict, cost_rate: float = 0.0015, min_c
                 order["buy"].pop(code)
         cost = max(min_cost, (cash_inflow + cash_outflow) * cost_rate)
         cash_needed = cash_outflow - cash_inflow + cost
-        # print("cash_needed: ", cash_needed, "cash: ", self.cash)
         if cash_needed > self.cash:
-            return order, False
-        else:
-            return order, True
+            # c_n = buy + r(buy + sell) - sell > c, 令n * buy + r(n * buy + sell) - sell = c
+            # 则n * buy * (1 + r) - (1 - r) * sell = c, 即 n * buy * (1 + r) = c + (1 - r)sell
+            # n = (c + (1 - r)sell) / buy(1 + r)
+            ratio = (self.cash + (1 - cost_rate) * cash_inflow) / (cash_outflow * (1 + cost_rate))
+            order["buy"] = {k: int(v * ratio + 0.5) for k, v in order["buy"].items()}  # 这样虽然不是整手下单, 但只要在1手以上都没问题
+        return order
 
     def update_price(self, price: dict):  # 更新市场价格
         for code in price.keys():

scutquant/alpha.py

@@ -19,12 +19,11 @@ def market_neutralize(x: pd.Series, long_only: bool = False) -> pd.Series:
     """
     _mean = x.groupby(level=0).mean()
     x -= _mean
-    if long_only:  # 考虑到A股有做空限制, 因此将权重为负的股票(即做空的股票)的权重调整为0(即纯多头), 并相应调整多头的权重
-        x[x.values < 0] = 0
-        abs_sum = x[x.values > 0].groupby(level=0).sum()
-    else:
-        abs_sum = abs(x).groupby(level=0).sum()
+    abs_sum = abs(x).groupby(level=0).sum()
     x /= abs_sum
+    if long_only:
+        x[x < 0] = 0
+        x *= 2
     return x
 
 
@@ -119,6 +118,7 @@ def call(self):
         pass
 
     def normalize(self):
+        self.result = mad_winsor(inf_mask(self.result))
         if self.norm_method == "zscore":
             self.result = cs_zscore(self.result)
         elif self.norm_method == "robust_zscore":

scutquant/executor.py

@@ -1,5 +1,43 @@
 from . import account, strategy
-from .signal_generator import *
+import pandas as pd
+import numpy as np
+import warnings
+
+warnings.filterwarnings("ignore")
+
+
+def get_daily_inter(data: pd.Series | pd.DataFrame, shuffle=False):
+    daily_count = data.groupby(level=0).size().values
+    daily_index = np.roll(np.cumsum(daily_count), 1)
+    daily_index[0] = 0
+    if shuffle:
+        daily_shuffle = list(zip(daily_index, daily_count))
+        np.random.shuffle(daily_shuffle)
+        daily_index, daily_count = zip(*daily_shuffle)
+    return daily_index, daily_count
+
+
+def prepare(predict: pd.DataFrame, data: pd.DataFrame, price: str, volume: str, real_ret: pd.Series) -> pd.DataFrame:
+    """
+    :param predict: pd.DataFrame, 预测值, 应包括"predict"
+    :param data: pd.DataFrame, 提供时间和价格信息
+    :param price: str, data中表示价格的列名
+    :param volume: str, data中表示成交量的列名
+    :param real_ret: pd.Series, 真实收益率
+    :return: pd.DataFrame
+    """
+    data_ = data.copy()
+    predict.columns = ["predict"]
+    index = predict.index
+    data1 = data_[data_.index.isin(index)]
+    data1 = data1.reset_index()
+    data1 = data1.set_index(predict.index.names).sort_index()
+    predict["price"] = data1[price]
+    predict["volume"] = data1[volume]  # 当天的交易量, 假设交易量不会发生大的跳跃
+    predict.index.names = ["time", "code"]
+    predict["price"] = predict["price"].groupby(["code"]).shift(-1)  # 指令是T时生成的, 但是T+1执行, 所以是shift(-1)
+    predict["R"] = real_ret[real_ret.index.isin(predict.index)]  # 本来就是T+2对T+1的收益率, 因此不用前移
+    return predict.dropna()
 
 
 class Executor:
@@ -61,18 +99,17 @@ def create_account(self):
         self.benchmark = account.Account(self.ben_cash, {}, {}, self.price.copy())
 
     def get_cash_available(self):
-        self.value_hold = 0.0
-        for code in self.user_account.price.keys():  # 更新持仓市值, 如果持有的资产在价格里面, 更新资产价值
-            if code in self.user_account.position.keys():
-                self.value_hold += self.user_account.position[code] * self.user_account.price[code]
-        # value是账户总价值, 乘risk_deg后得到所有可交易资金, 减去value_hold就是剩余可交易资金
-        return self.user_account.value * self.s.risk_degree - self.value_hold
+        """
+        fixme: 调整计算方式使其适应先卖后买的情况
+        之所以不用cash * risk_degree是因为先卖后买的情况下, 当前的cash跟实际可支配的cash不一样(因为卖了就有钱了)
+        """
+        return self.user_account.value * self.s.risk_degree
 
     def execute(self, data: pd.DataFrame, verbose: int = 0):
         """
         :param data: pd.DataFrame, 包括三列：'predict', 'volume', 'price', 'label' 以及多重索引[('time', 'code')]
         :param verbose: int, 是否输出交易记录
-        :return: self
+        :return:
         """
 
         def check_names(index=data.index, predict="predict", price="price"):
@@ -88,26 +125,31 @@ def check_names(index=data.index, predict="predict", price="price"):
         self.init_account(data)
         self.create_account()
         if self.mode == "generate":
-            time = data.index.get_level_values(0).unique().values
-            for t in time:
-                idx = data["R"].groupby(data.index.names[0]).mean()  # 大盘收益率
-                self.time.append(t)
-                data_select = data[data.index.get_level_values(0) == t]
-                signal = generate(data=data_select, strategy=self.s, cash_available=self.get_cash_available())
-                order, current_price = signal["order"], signal["current_price"]
+            benchmark = data["R"].groupby(level=0).transform(lambda x: x.mean())  # 大盘收益率
+            daily_idx, daily_count = get_daily_inter(data)
+            for idx, count in zip(daily_idx, daily_count):
+                batch = slice(idx, idx + count)
+                data_batch = data.iloc[batch]
+                benchmark_batch = benchmark.iloc[batch]
+                current_day = data_batch.index.get_level_values(0)[0]
+                self.time.append(current_day)
+                order, current_price = self.s.to_signal(data_batch, position=self.user_account.position,
+                                                        cash_available=self.get_cash_available())
 
                 if self.s.auto_offset:
-                    order = self.user_account.generate_total_order(order=order, freq=self.s.offset_freq)
-                order, trade = self.user_account.check_order(order, current_price)
-
-                if verbose == 1 and trade:
-                    print(t, '\n', "buy:", '\n', order["buy"], '\n', "sell:", order["sell"], '\n')
+                    order = self.user_account.adjust_order(order=order, freq=self.s.offset_freq)
+                order = self.user_account.check_order(order, current_price)
+                # print(trade)
+                # trade = True
+                if verbose == 1:
+                    print(current_day, '\n', "buy:", '\n', order["buy"], '\n', "sell:", order["sell"], '\n')
 
                 self.user_account.update_all(order=order, price=current_price, cost_buy=self.cost_buy,
-                                             cost_sell=self.cost_sell, min_cost=self.min_cost, trade=trade)
+                                             cost_sell=self.cost_sell, min_cost=self.min_cost)
                 self.user_account.risk_control(risk_degree=self.s.risk_degree, cost_rate=self.cost_sell,
                                                min_cost=self.min_cost)
-                self.benchmark.value *= (1 + idx[idx.index == t][0])  # 乘上1+大盘收益率, 相当于等权指数
+
+                self.benchmark.value *= 1 + benchmark_batch.values[0]  # 乘上1+大盘收益率, 相当于等权指数
                 self.benchmark.val_hist.append(self.benchmark.value)
         else:
             raise ValueError("simulate mode is not available by far")

scutquant/report.py

@@ -82,9 +82,15 @@ def plot(data, label, title: str = None, xlabel: str = None, ylabel: str = None,
         plt.figure(figsize=figsize)
     # plt.clf()
     if mode == "plot":
-        for d in range(len(data)):
-            plt.plot(data[d], label=label[d])
-            plt.xticks(rotation=45)
+        if len(data) > 10:
+            cmap = plt.get_cmap('tab20')
+            for d in range(len(data)):
+                plt.plot(data[d], label=label[d], color=cmap(d))
+                plt.xticks(rotation=45)
+        else:
+            for d in range(len(data)):
+                plt.plot(data[d], label=label[d])
+                plt.xticks(rotation=45)
     elif mode == "bar":
         bar = plt.bar(label, data, label="value")
         plt.bar_label(bar, label_type='edge')
@@ -122,16 +128,13 @@ def accuracy(pred: pd.Series, y: pd.Series, sign: str = ">=") -> float:
     return len(data_true) / len(data)
 
 
-def report_all(user_account, benchmark, show_raw_value: bool = False, excess_return: bool = True, risk: bool = True,
-               turnover: bool = True, rf: float = 0.03, freq: float = 1, time=None, figsize: tuple = (10, 6)) -> None:
+def report_all(user_account, benchmark, show_raw_value: bool = False, rf: float = 0.03, freq: float = 1, time=None,
+               figsize: tuple = (10, 6)) -> None:
     """
 
     :param user_account: account类
     :param benchmark: account类
     :param show_raw_value: 显示原始市值（具体金额）
-    :param excess_return: 显示超额收益曲线
-    :param risk: 显示风险度
-    :param turnover: 显示换手率
     :param rf: 显示无风险利率
     :param freq: 频率, 日频为1，月频为30，其它类推
     :param time: 显示时间轴
@@ -159,20 +162,22 @@ def report_all(user_account, benchmark, show_raw_value: bool = False, excess_ret
             days += 1
     days /= len(acc_ret)
 
-    acc_dd = calc_drawdown(pd.Series(acc_ret))
-    ben_dd = calc_drawdown(pd.Series(ben_ret))
+    acc_ret = pd.Series(acc_ret, name="acc_ret", index=time)  # 累计收益率
+    ben_ret = pd.Series(ben_ret, name="ben_ret", index=time)  # benchmark的累计收益率
+    excess_ret = pd.Series(excess_ret, name="excess_ret", index=time)
 
-    ret = pd.Series(acc_ret)  # 累计收益率
-    ben = pd.Series(ben_ret)  # benchmark的累计收益率
+    acc_dd = calc_drawdown(acc_ret)
+    ben_dd = calc_drawdown(ben_ret)
+    excess_dd = calc_drawdown(excess_ret)
 
-    ann_return = annualized_return(ret, freq=freq)
-    ann_std = annualized_volatility(ret, freq=freq)
-    ben_ann_return = annualized_return(ben, freq=freq)
-    ben_ann_std = annualized_volatility(ben, freq=freq)
+    ann_return = annualized_return(acc_ret, freq=freq)
+    ann_std = annualized_volatility(acc_ret, freq=freq)
+    ben_ann_return = annualized_return(ben_ret, freq=freq)
+    ben_ann_std = annualized_volatility(ben_ret, freq=freq)
 
-    beta = ret.cov(ben) / ben.var()
-    alpha = ret.mean() - beta * ben.mean()
-    epsilon = pd.Series(ret - beta * ben - alpha).std()
+    beta = acc_ret.cov(ben_ret) / ben_ret.var()
+    alpha = acc_ret.mean() - beta * ben_ret.mean()
+    epsilon = (acc_ret - beta * ben_ret - alpha).std()
 
     sharpe = sharpe_ratio(acc_ret, rf=rf, freq=freq * 365)
     sortino = sortino_ratio(acc_ret, ben_ret)
@@ -207,10 +212,6 @@ def report_all(user_account, benchmark, show_raw_value: bool = False, excess_ret
         plt.legend()
         plt.show()
     else:
-        acc_ret = pd.Series(acc_ret, name="acc_ret", index=time)
-        ben_ret = pd.Series(ben_ret, name="ben_ret", index=time)
-        excess_ret = pd.Series(excess_ret, name="excess_ret", index=time)
-
         plt.figure(figsize=(10, 6))
         plt.plot(acc_ret, label="return", color="red")
         plt.plot(ben_ret, label="benchmark", color="blue")
@@ -222,21 +223,19 @@ def report_all(user_account, benchmark, show_raw_value: bool = False, excess_ret
     plt.clf()
     plt.figure(figsize=(10, 6))
     plt.plot(acc_dd, label="drawdown")
-    plt.plot(ben_dd, label="excess_return_drawdown")
+    plt.plot(excess_dd, label="excess_return_drawdown")
     plt.legend()
     plt.title("Drawdown")
     plt.show()
 
-    if risk:
-        risk = pd.DataFrame({'risk': user_account.risk_curve}, index=time)
-        plot([risk], label=['risk_degree'], title='Risk Degree', ylabel='value', figsize=figsize)
+    risk = pd.DataFrame({'risk': user_account.risk_curve}, index=time)
+    plot([risk], label=['risk_degree'], title='Risk Degree', ylabel='value', figsize=figsize)
 
-    if turnover:
-        risk = pd.DataFrame({'turnover': user_account.turnover}, index=time)
-        plot([risk], label=['turnover'], title='Turnover', figsize=figsize)
+    risk = pd.DataFrame({'turnover': user_account.turnover}, index=time)
+    plot([risk], label=['turnover'], title='Turnover', figsize=figsize)
 
 
-def group_return_ana(pred: pd.DataFrame | pd.Series, y_true: pd.Series, n: int = 5, groupby: str = "time",
+def group_return_ana(pred: pd.DataFrame | pd.Series, y_true: pd.Series, n: int = 10, groupby: str = "time",
                      figsize: tuple = (10, 6)) -> None:
     """
     因子对股票是否有良好的区分度, 若有, 则应出现明显的分层效应(即单调性)
@@ -283,12 +282,10 @@ def group_return_ana(pred: pd.DataFrame | pd.Series, y_true: pd.Series, n: int =
     win_rate = []
     mean_ret = []
     for c in cols:
-        # dt = t_df[c] + 1
-        # data.append(dt.cumprod() - 1)
         data.append(t_df[c].cumsum())
         label.append(c)
-        win_rate.append(len(t_df[t_df[c] >= 0]) / len(t_df))
-        mean_ret.append(t_df[c].cumsum().values[-1] / len(t_df) * 100)
+        win_rate.append(round(len(t_df[t_df[c] >= 0]) / len(t_df), 4))
+        mean_ret.append(round(t_df[c].cumsum().values[-1] / len(t_df) * 100, 4))
     plot(data, label, title='Grouped Return', xlabel='time_id', ylabel='value', figsize=figsize)
     plot(win_rate, label=cols, title="Win Rate of Each Group", mode="bar", figsize=figsize)
     plot(mean_ret, label=cols, title="Mean Return of Each Group(%)", mode="bar", figsize=figsize)