feat: 新增因子装饰器系统和完整因子文档
- 添加因子表达式文档，收录180+个因子及数学表达式 - 添加因子实现分析报告，明确ts_*与cs_*算子分类 - 实现装饰器系统：@time_series/@cross_section/@element_wise - 优化API和翻译器以支持新架构
This commit is contained in:
615
docs/factor_expressions_document.md
Normal file
615
docs/factor_expressions_document.md
Normal file
@@ -0,0 +1,615 @@
|
||||
# 因子名称与表达式文档
|
||||
|
||||
## 数据来源
|
||||
- 分析文件: `main/train/Classify2_load_model.ipynb`
|
||||
- 因子模块: `main/factor/factor.py`, `main/factor/money_factor.py`, `main/factor/utils.py`
|
||||
|
||||
---
|
||||
|
||||
## 一、财务因子 (Financial Factors)
|
||||
|
||||
### 1. add_financial_factor 系列
|
||||
**因子名称**: `undist_profit_ps`, `ocfps`, `roa`, `roe`
|
||||
|
||||
**表达式**:
|
||||
- 使用 `merge_asof` 将财务指标数据按股票代码和公告日期匹配到每个交易日
|
||||
- 匹配逻辑: 向后查找（找 ≤ trade_date 的最近财务数据）
|
||||
- 公式: `factor_value` 直接作为因子值
|
||||
|
||||
### 2. calculate_cashflow_to_ev_factor
|
||||
**因子名称**: `cashflow_to_ev_factor`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
Enterprise Value = total_mv * 10000 + total_liab - money_cap
|
||||
cashflow_to_ev_factor = n_cashflow_act / Enterprise Value
|
||||
```
|
||||
|
||||
### 3. caculate_book_to_price_ratio
|
||||
**因子名称**: `book_to_price_ratio`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
book_to_price_ratio = bps / close
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 二、ARBR 因子 (ARBR Factors)
|
||||
|
||||
### 4. calculate_arbr
|
||||
**因子名称**: `AR`, `BR`, `AR_BR`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
# 中间计算
|
||||
h_minus_o = high - open
|
||||
o_minus_l = open - low
|
||||
prev_close = close.shift(1)
|
||||
h_minus_pc_pos = max(0, high - prev_close)
|
||||
pc_minus_l_pos = max(0, prev_close - low)
|
||||
|
||||
# AR 和 BR 计算
|
||||
AR = sum(h_minus_o, N) / sum(o_minus_l, N) * 100
|
||||
BR = sum(h_minus_pc_pos, N) / sum(pc_minus_l_pos, N) * 100
|
||||
AR_BR = AR - BR
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 三、技术指标因子 (Technical Indicator Factors)
|
||||
|
||||
### 5. turnover_rate_n
|
||||
**因子名称**: `turnover_rate_mean_5`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
turnover_rate_mean_5 = mean(turnover_rate, window=5)
|
||||
```
|
||||
|
||||
### 6. variance_n
|
||||
**因子名称**: `variance_20`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
variance_20 = var(pct_chg, window=20)
|
||||
```
|
||||
|
||||
### 7. bbi_ratio_factor
|
||||
**因子名称**: `bbi_ratio_factor`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
SMA3 = mean(close, 3)
|
||||
SMA6 = mean(close, 6)
|
||||
SMA12 = mean(close, 12)
|
||||
SMA24 = mean(close, 24)
|
||||
BBI = (SMA3 + SMA6 + SMA12 + SMA24) / 4
|
||||
bbi_ratio_factor = BBI / close
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 四、偏离度因子 (Deviation Factors)
|
||||
|
||||
### 8. daily_deviation
|
||||
**因子名称**: `daily_deviation`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
# 计算日级别动量基准
|
||||
daily_positive_benchmark = mean(pct_chg[pct_chg > 0]) # 每日上涨股票的平均涨跌幅
|
||||
daily_negative_benchmark = mean(pct_chg[pct_chg < 0]) # 每日下跌股票的平均涨跌幅
|
||||
|
||||
# 偏离度计算
|
||||
if pct_chg > 0 and daily_positive_benchmark > 0:
|
||||
daily_deviation = pct_chg - daily_positive_benchmark
|
||||
elif pct_chg < 0 and daily_negative_benchmark < 0:
|
||||
daily_deviation = pct_chg - daily_negative_benchmark
|
||||
else:
|
||||
daily_deviation = 0
|
||||
```
|
||||
|
||||
### 9. daily_industry_deviation
|
||||
**因子名称**: `daily_industry_deviation`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
# 计算日级别行业动量基准
|
||||
daily_industry_positive_benchmark = mean(pct_chg[pct_chg > 0]) # 按 trade_date + cat_l2_code 分组
|
||||
daily_industry_negative_benchmark = mean(pct_chg[pct_chg < 0]) # 按 trade_date + cat_l2_code 分组
|
||||
|
||||
# 行业偏离度计算
|
||||
if pct_chg > 0 and daily_industry_positive_benchmark > 0:
|
||||
daily_industry_deviation = pct_chg - daily_industry_positive_benchmark
|
||||
elif pct_chg < 0 and daily_industry_negative_benchmark < 0:
|
||||
daily_industry_deviation = pct_chg - daily_industry_negative_benchmark
|
||||
else:
|
||||
daily_industry_deviation = 0
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 五、滚动因子和简单因子 (Rolling & Simple Factors)
|
||||
|
||||
### 10. get_rolling_factor 生成的因子
|
||||
|
||||
#### 资金流因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `lg_elg_net_buy_vol` | `(buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol)` |
|
||||
| `flow_lg_elg_intensity` | `lg_elg_net_buy_vol / (vol + epsilon)` |
|
||||
| `sm_net_buy_vol` | `buy_sm_vol - sell_sm_vol` |
|
||||
| `flow_divergence_diff` | `sm_net_buy_vol - lg_elg_net_buy_vol` |
|
||||
| `flow_divergence_ratio` | `sm_net_buy_vol / (lg_elg_net_buy_vol + sign(lg_elg_net_buy_vol) * epsilon + epsilon)` |
|
||||
| `total_buy_vol` | `buy_sm_vol + buy_lg_vol + buy_elg_vol` |
|
||||
| `lg_elg_buy_prop` | `(buy_lg_vol + buy_elg_vol) / (total_buy_vol + epsilon)` |
|
||||
| `flow_struct_buy_change` | `diff(lg_elg_buy_prop, 1)` |
|
||||
| `lg_elg_net_buy_vol_change` | `diff(lg_elg_net_buy_vol, 1)` |
|
||||
| `flow_lg_elg_accel` | `diff(lg_elg_net_buy_vol_change, 1)` |
|
||||
|
||||
#### 筹码分布因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `chip_concentration_range` | `(cost_95pct - cost_5pct) / (close + epsilon)` |
|
||||
| `chip_skewness` | `(weight_avg - cost_50pct) / (cost_50pct + epsilon)` |
|
||||
| `floating_chip_proxy` | `winner_rate * max(0, (close - cost_15pct) / (close + epsilon))` |
|
||||
| `cost_support_15pct_change` | `pct_change(cost_15pct, 1) * 100` |
|
||||
| `cat_winner_price_zone` | categorical: 1=高风险区, 2=低潜力区, 3=中上获利区, 4=中下亏损区 |
|
||||
| `flow_chip_consistency` | `lg_elg_net_buy_vol * price_near_low_support` |
|
||||
| `profit_taking_vs_absorb` | `lg_elg_net_buy_vol * (winner_rate > 0.7)` |
|
||||
|
||||
#### 波动率因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `upside_vol` | `std(pos_returns, window=20)` |
|
||||
| `downside_vol` | `std(neg_returns, window=20)` |
|
||||
| `vol_ratio` | `upside_vol / downside_vol` |
|
||||
| `return_skew` | `skew(pct_chg, window=5)` |
|
||||
| `return_kurtosis` | `kurt(pct_chg, window=5)` |
|
||||
|
||||
#### 成交量因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `volume_change_rate` | `mean(vol, 2) / mean(vol, 10) - 1` |
|
||||
| `cat_volume_breakout` | `vol > max(vol, 5)` |
|
||||
| `turnover_deviation` | `(turnover_rate - mean(turnover_rate, 3)) / std(turnover_rate, 3)` |
|
||||
| `cat_turnover_spike` | `turnover_rate > mean(turnover_rate, 3) + 2 * std(turnover_rate, 3)` |
|
||||
| `avg_volume_ratio` | `mean(volume_ratio, 3)` |
|
||||
| `cat_volume_ratio_breakout` | `volume_ratio > max(volume_ratio, 5)` |
|
||||
| `vol_spike` | `mean(vol, 20)` |
|
||||
| `vol_std_5` | `std(pct_change(vol), 5)` |
|
||||
|
||||
#### 技术指标
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `atr_14` | `ATR(high, low, close, 14)` (TA-Lib) |
|
||||
| `atr_6` | `ATR(high, low, close, 6)` (TA-Lib) |
|
||||
| `obv` | `OBV(close, vol)` (TA-Lib) |
|
||||
| `maobv_6` | `SMA(obv, 6)` (TA-Lib) |
|
||||
| `rsi_3` | `RSI(close, 3)` (TA-Lib) |
|
||||
|
||||
#### 收益率因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `return_5` | `close / close.shift(5) - 1` |
|
||||
| `return_20` | `close / close.shift(20) - 1` |
|
||||
| `std_return_5` | `std(pct_change(close), 5)` |
|
||||
| `std_return_90` | `std(pct_change(close), 90)` |
|
||||
| `std_return_90_2` | `std(pct_change(close.shift(10)), 90)` |
|
||||
|
||||
#### EMA 因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `act_factor1` | `atan((EMA(close,5)/EMA(close,5).shift(1)-1)*100) * 57.3 / 50` |
|
||||
| `act_factor2` | `atan((EMA(close,13)/EMA(close,13).shift(1)-1)*100) * 57.3 / 40` |
|
||||
| `act_factor3` | `atan((EMA(close,20)/EMA(close,20).shift(1)-1)*100) * 57.3 / 21` |
|
||||
| `act_factor4` | `atan((EMA(close,60)/EMA(close,60).shift(1)-1)*100) * 57.3 / 10` |
|
||||
| `rank_act_factor1` | `rank(act_factor1, pct=True)` |
|
||||
| `rank_act_factor2` | `rank(act_factor2, pct=True)` |
|
||||
| `rank_act_factor3` | `rank(act_factor3, pct=True)` |
|
||||
| `log_circ_mv` | `log(circ_mv)` |
|
||||
|
||||
#### Alpha 因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `cov` | `cov(high, vol, window=5)` |
|
||||
| `delta_cov` | `diff(cov, 5)` |
|
||||
| `_stddev_close` | `std(close, 20)` |
|
||||
| `_rank_stddev` | `rank(_stddev_close, pct=True)` |
|
||||
| `alpha_22_improved` | `-1 * delta_cov * _rank_stddev` |
|
||||
| `alpha_003` | `(close - open) / (high - low)` (if high != low else 0) |
|
||||
| `alpha_007` | `rank(rolling_corr(close, vol, 5), pct=True)` |
|
||||
| `alpha_013` | `rank(sum(close, 5) - sum(close, 20), pct=True)` |
|
||||
|
||||
#### 筹码因子
|
||||
| 因子名称 | 表达式 |
|
||||
|---------|--------|
|
||||
| `vol_break` | `1 if (close > cost_85pct) & (volume_ratio > 2) else 0` |
|
||||
| `weight_roc5` | `pct_change(weight_avg, 5)` |
|
||||
| `price_cost_divergence` | `corr(pct_change(close), pct_change(weight_avg), 10)` |
|
||||
| `smallcap_concentration` | `(1 / log_circ_mv) * (cost_85pct - cost_15pct)` |
|
||||
| `cost_stability` | `std(weight_avg, 20) / mean(weight_avg, 20)` |
|
||||
| `high_cost_break_days` | `sum(close > cost_95pct, 5)` |
|
||||
| `liquidity_risk` | `(cost_95pct - cost_5pct) / mean(vol, 10)` |
|
||||
| `turnover_std` | `std(turnover_rate, 20)` |
|
||||
| `mv_volatility` | `turnover_std / log_circ_mv` |
|
||||
| `volume_growth` | `pct_change(vol, 20)` |
|
||||
| `mv_growth` | `volume_growth / log_circ_mv` |
|
||||
| `momentum_factor` | `volume_change_rate + 0.5 * turnover_deviation` |
|
||||
| `resonance_factor` | `vol_ratio * pct_chg` |
|
||||
| `log_close` | `log(close)` |
|
||||
| `cat_vol_spike` | `vol > 2 * vol_spike` |
|
||||
| `up` | `(high - max(close, open)) / close` |
|
||||
| `down` | `(min(close, open) - low) / close` |
|
||||
| `obv_maobv_6` | `obv - maobv_6` |
|
||||
| `std_return_5_over_std_return_90` | `std_return_5 / std_return_90` |
|
||||
| `std_return_90_minus_std_return_90_2` | `std_return_90 - std_return_90_2` |
|
||||
| `cat_af2` | `act_factor2 > act_factor1` |
|
||||
| `cat_af3` | `act_factor3 > act_factor2` |
|
||||
| `cat_af4` | `act_factor4 > act_factor3` |
|
||||
| `act_factor5` | `act_factor1 + act_factor2 + act_factor3 + act_factor4` |
|
||||
| `act_factor6` | `(act_factor1 - act_factor2) / sqrt(act_factor1^2 + act_factor2^2)` |
|
||||
| `active_buy_volume_large` | `buy_lg_vol / net_mf_vol` |
|
||||
| `active_buy_volume_big` | `buy_elg_vol / net_mf_vol` |
|
||||
| `active_buy_volume_small` | `buy_sm_vol / net_mf_vol` |
|
||||
| `buy_lg_vol_minus_sell_lg_vol` | `(buy_lg_vol - sell_lg_vol) / net_mf_vol` |
|
||||
| `buy_elg_vol_minus_sell_elg_vol` | `(buy_elg_vol - sell_elg_vol) / net_mf_vol` |
|
||||
| `ctrl_strength` | `(cost_85pct - cost_15pct) / (his_high - his_low)` |
|
||||
| `low_cost_dev` | `(close - cost_5pct) / (cost_50pct - cost_5pct)` |
|
||||
| `asymmetry` | `(cost_95pct - cost_50pct) / (cost_50pct - cost_5pct)` |
|
||||
| `lock_factor` | `turnover_rate * (1 - (cost_95pct - cost_5pct) / (his_high - his_low))` |
|
||||
| `cat_vol_break` | `(close > cost_85pct) & (volume_ratio > 2)` |
|
||||
| `cost_atr_adj` | `(cost_95pct - cost_5pct) / atr_14` |
|
||||
| `cat_golden_resonance` | `(close > weight_avg) & (volume_ratio > 1.5) & (winner_rate > 0.7)` |
|
||||
| `mv_turnover_ratio` | `turnover_rate / log_circ_mv` |
|
||||
| `mv_adjusted_volume` | `vol / log_circ_mv` |
|
||||
| `mv_weighted_turnover` | `turnover_rate / log_circ_mv` |
|
||||
| `nonlinear_mv_volume` | `vol / log_circ_mv` |
|
||||
| `mv_volume_ratio` | `volume_ratio / log_circ_mv` |
|
||||
| `mv_momentum` | `turnover_rate * volume_ratio / log_circ_mv` |
|
||||
|
||||
---
|
||||
|
||||
## 六、资金流因子 (Money Flow Factors)
|
||||
|
||||
### 11. lg_flow_mom_corr_20_60
|
||||
**表达式**:
|
||||
```
|
||||
net_lg_flow_val = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) * close
|
||||
rolling_net_lg_flow = sum(net_lg_flow_val, 20)
|
||||
price_mom = pct_change(close, 20)
|
||||
lg_flow_mom_corr_20_60 = corr(rolling_net_lg_flow, price_mom, 60)
|
||||
```
|
||||
|
||||
### 12. lg_flow_accel
|
||||
**表达式**:
|
||||
```
|
||||
net_lg_flow_vol = buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol
|
||||
lg_flow_accel = diff(diff(net_lg_flow_vol, 1), 1)
|
||||
```
|
||||
|
||||
### 13. profit_pressure
|
||||
**表达式**:
|
||||
```
|
||||
profit_margin_85 = close / cost_85pct - 1
|
||||
profit_margin_95 = close / cost_95pct - 1
|
||||
profit_pressure = winner_rate * 0.5 * (profit_margin_85 + profit_margin_95)
|
||||
```
|
||||
|
||||
### 14. underwater_resistance
|
||||
**表达式**:
|
||||
```
|
||||
underwater_ratio = 1.0 - winner_rate
|
||||
dist_to_cost_15 = max(0, cost_15pct - close) / (close + epsilon)
|
||||
underwater_resistance = underwater_ratio * dist_to_cost_15
|
||||
```
|
||||
|
||||
### 15. cost_conc_std_20
|
||||
**表达式**:
|
||||
```
|
||||
cost_range_norm = (cost_85pct - cost_15pct) / (weight_avg + epsilon)
|
||||
cost_conc_std_20 = std(cost_range_norm, 20)
|
||||
```
|
||||
|
||||
### 16. profit_decay_20
|
||||
**表达式**:
|
||||
```
|
||||
ret_20 = close / close.shift(20) - 1
|
||||
winner_rate_change_20 = diff(winner_rate, 20)
|
||||
profit_decay_20 = ret_20 / winner_rate_change_20
|
||||
```
|
||||
|
||||
### 17. vol_amp_loss_20
|
||||
**表达式**:
|
||||
```
|
||||
vol_20 = std(pct_chg, 20)
|
||||
loss_degree = max(0, weight_avg - close) / (close + epsilon)
|
||||
vol_amp_loss_20 = vol_20 * loss_degree
|
||||
```
|
||||
|
||||
### 18. vol_drop_profit_cnt_5
|
||||
**表达式**:
|
||||
```
|
||||
is_profitable = close > weight_avg * (1 + 0.1)
|
||||
is_dropping = pct_chg < -0.03
|
||||
rolling_mean_vol = mean(vol, 20)
|
||||
rolling_std_vol = std(vol, 20)
|
||||
is_high_vol = vol > (rolling_mean_vol + 2 * rolling_std_vol)
|
||||
event = is_profitable & is_dropping & is_high_vol
|
||||
vol_drop_profit_cnt_5 = sum(event, 5)
|
||||
```
|
||||
|
||||
### 19. lg_flow_vol_interact_20
|
||||
**表达式**:
|
||||
```
|
||||
vol_20 = std(pct_chg, 20)
|
||||
net_lg_flow_val = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) * close
|
||||
total_val = vol * close
|
||||
abs_net_lg_flow_ratio = abs(net_lg_flow_val) / (total_val + epsilon)
|
||||
abs_net_lg_flow_ratio_20 = mean(abs_net_lg_flow_ratio, 20)
|
||||
lg_flow_vol_interact_20 = vol_20 * abs_net_lg_flow_ratio_20
|
||||
```
|
||||
|
||||
### 20. cost_break_confirm_cnt_5
|
||||
**表达式**:
|
||||
```
|
||||
prev_cost_85 = cost_85pct.shift(1)
|
||||
prev_cost_15 = cost_15pct.shift(1)
|
||||
break_up = close > prev_cost_85
|
||||
break_down = close < prev_cost_15
|
||||
net_lg_flow_vol = buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol
|
||||
confirm_up = break_up & (net_lg_flow_vol > 0)
|
||||
confirm_down = break_down & (net_lg_flow_vol < 0)
|
||||
net_confirm = confirm_up - confirm_down
|
||||
cost_break_confirm_cnt_5 = sum(net_confirm, 5)
|
||||
```
|
||||
|
||||
### 21. atr_norm_channel_pos_14
|
||||
**表达式**:
|
||||
```
|
||||
tr = max(high - low, abs(high - prev_close), abs(low - prev_close))
|
||||
atr_14 = mean(tr, 14)
|
||||
roll_low_14 = min(low, 14)
|
||||
atr_norm_channel_pos_14 = (close - roll_low_14) / atr_14
|
||||
```
|
||||
|
||||
### 22. turnover_diff_skew_20
|
||||
**表达式**:
|
||||
```
|
||||
turnover_diff = diff(turnover_rate, 1)
|
||||
turnover_diff_skew_20 = skew(turnover_diff, 20)
|
||||
```
|
||||
|
||||
### 23. lg_sm_flow_diverge_20
|
||||
**表达式**:
|
||||
```
|
||||
lg_flow_ratio = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) / vol
|
||||
sm_flow_ratio = (buy_sm_vol - sell_sm_vol) / vol
|
||||
lg_flow_ratio_20 = mean(lg_flow_ratio, 20)
|
||||
sm_flow_ratio_20 = mean(sm_flow_ratio, 20)
|
||||
lg_sm_flow_diverge_20 = lg_flow_ratio_20 - sm_flow_ratio_20
|
||||
```
|
||||
|
||||
### 24. pullback_strong_20_20
|
||||
**表达式**:
|
||||
```
|
||||
high_20 = max(high, 20)
|
||||
pullback_depth = (high_20 - close) / high_20
|
||||
recent_gain_20 = close / close.shift(20) - 1
|
||||
pullback_strong_20_20 = pullback_depth / recent_gain_20
|
||||
```
|
||||
|
||||
### 25. vol_wgt_hist_pos_20
|
||||
**表达式**:
|
||||
```
|
||||
hist_pos = (close - his_low) / (his_high - his_low)
|
||||
rolling_mean_vol_20 = mean(vol, 20)
|
||||
vol_rel_strength = vol / rolling_mean_vol_20
|
||||
vol_wgt_hist_pos_20 = hist_pos * vol_rel_strength
|
||||
```
|
||||
|
||||
### 26. vol_adj_roc_20
|
||||
**表达式**:
|
||||
```
|
||||
roc_20 = close / close.shift(20) - 1
|
||||
vol_20 = std(pct_chg, 20)
|
||||
vol_adj_roc_20 = roc_20 / vol_20
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 七、截面排序因子 (Cross-Sectional Rank Factors)
|
||||
|
||||
### 27. cs_rank_net_lg_flow_val
|
||||
**表达式**:
|
||||
```
|
||||
net_lg_flow_val = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) * close
|
||||
cs_rank_net_lg_flow_val = rank(net_lg_flow_val, pct=True)
|
||||
```
|
||||
|
||||
### 28. cs_rank_flow_divergence
|
||||
**表达式**:
|
||||
```
|
||||
lg_ratio = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) / vol
|
||||
sm_ratio = (buy_sm_vol - sell_sm_vol) / vol
|
||||
divergence = lg_ratio - sm_ratio
|
||||
cs_rank_flow_divergence = rank(divergence, pct=True)
|
||||
```
|
||||
|
||||
### 29. cs_rank_ind_adj_lg_flow
|
||||
**表达式**:
|
||||
```
|
||||
net_lg_flow_vol = (buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol) * close
|
||||
industry_avg_flow = mean(net_lg_flow_vol) by trade_date, cat_l2_code
|
||||
deviation = net_lg_flow_vol - industry_avg_flow
|
||||
cs_rank_ind_adj_lg_flow = rank(deviation, pct=True)
|
||||
```
|
||||
|
||||
### 30. cs_rank_elg_buy_ratio
|
||||
**表达式**:
|
||||
```
|
||||
elg_buy_ratio = buy_elg_vol / vol
|
||||
cs_rank_elg_buy_ratio = rank(elg_buy_ratio, pct=True)
|
||||
```
|
||||
|
||||
### 31. cs_rank_rel_profit_margin
|
||||
**表达式**:
|
||||
```
|
||||
profit_margin = (close - weight_avg) / close
|
||||
cs_rank_rel_profit_margin = rank(profit_margin, pct=True)
|
||||
```
|
||||
|
||||
### 32. cs_rank_cost_breadth
|
||||
**表达式**:
|
||||
```
|
||||
cost_breadth = (cost_85pct - cost_15pct) / weight_avg
|
||||
cs_rank_cost_breadth = rank(cost_breadth, pct=True)
|
||||
```
|
||||
|
||||
### 33. cs_rank_dist_to_upper_cost
|
||||
**表达式**:
|
||||
```
|
||||
dist_to_95 = close / cost_95pct
|
||||
cs_rank_dist_to_upper_cost = rank(dist_to_95, pct=True)
|
||||
```
|
||||
|
||||
### 34. cs_rank_winner_rate
|
||||
**表达式**:
|
||||
```
|
||||
cs_rank_winner_rate = rank(winner_rate, pct=True)
|
||||
```
|
||||
|
||||
### 35. cs_rank_intraday_range
|
||||
**表达式**:
|
||||
```
|
||||
norm_range = (high - low) / close
|
||||
cs_rank_intraday_range = rank(norm_range, pct=True)
|
||||
```
|
||||
|
||||
### 36. cs_rank_close_pos_in_range
|
||||
**表达式**:
|
||||
```
|
||||
close_pos = (close - low) / (high - low)
|
||||
cs_rank_close_pos_in_range = rank(close_pos, pct=True)
|
||||
```
|
||||
|
||||
### 37. cs_rank_opening_gap
|
||||
**表达式**:
|
||||
```
|
||||
gap = open / pre_close - 1
|
||||
cs_rank_opening_gap = rank(gap, pct=True)
|
||||
```
|
||||
|
||||
### 38. cs_rank_pos_in_hist_range
|
||||
**表达式**:
|
||||
```
|
||||
hist_pos = (close - his_low) / (his_high - his_low)
|
||||
cs_rank_pos_in_hist_range = rank(hist_pos, pct=True)
|
||||
```
|
||||
|
||||
### 39. cs_rank_vol_x_profit_margin
|
||||
**表达式**:
|
||||
```
|
||||
daily_vol = abs(pct_chg)
|
||||
profit_margin = (close - weight_avg) / close
|
||||
interaction = daily_vol * profit_margin
|
||||
cs_rank_vol_x_profit_margin = rank(interaction, pct=True)
|
||||
```
|
||||
|
||||
### 40. cs_rank_lg_flow_price_concordance
|
||||
**表达式**:
|
||||
```
|
||||
net_lg_flow_vol = buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol
|
||||
concordance = net_lg_flow_vol * pct_chg
|
||||
cs_rank_lg_flow_price_concordance = rank(concordance, pct=True)
|
||||
```
|
||||
|
||||
### 41. cs_rank_turnover_per_winner
|
||||
**表达式**:
|
||||
```
|
||||
turnover_per_winner = turnover_rate / winner_rate
|
||||
cs_rank_turnover_per_winner = rank(turnover_per_winner, pct=True)
|
||||
```
|
||||
|
||||
### 42. cs_rank_ind_cap_neutral_pe
|
||||
**表达式**: `Placeholder - 需要 statsmodels 实现`
|
||||
|
||||
### 43. cs_rank_volume_ratio
|
||||
**表达式**:
|
||||
```
|
||||
cs_rank_volume_ratio = rank(volume_ratio, pct=True)
|
||||
```
|
||||
|
||||
### 44. cs_rank_elg_buy_sell_sm_ratio
|
||||
**表达式**:
|
||||
```
|
||||
ratio = buy_elg_vol / sell_sm_vol
|
||||
cs_rank_elg_buy_sell_sm_ratio = rank(ratio, pct=True)
|
||||
```
|
||||
|
||||
### 45. cs_rank_cost_dist_vol_ratio
|
||||
**表达式**:
|
||||
```
|
||||
dist = abs(close - weight_avg) / (close + epsilon)
|
||||
interaction = dist * volume_ratio
|
||||
cs_rank_cost_dist_vol_ratio = rank(interaction, pct=True)
|
||||
```
|
||||
|
||||
### 46. cs_rank_size
|
||||
**表达式**:
|
||||
```
|
||||
log_circ_mv = log1p(circ_mv)
|
||||
cs_rank_size = rank(log_circ_mv, pct=True)
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 八、行业因子 (Industry Factors)
|
||||
|
||||
### 47. get_act_factor (from main.utils.factor)
|
||||
**生成的因子**: `act_factor1`, `act_factor2`, `act_factor3`, `act_factor4`
|
||||
|
||||
**表达式**:
|
||||
```
|
||||
obv = OBV(close, vol)
|
||||
return_5 = close / close.shift(5) - 1
|
||||
return_20 = close / close.shift(20) - 1
|
||||
return_5_percentile = rank(return_5, pct=True)
|
||||
return_20_percentile = rank(return_20, pct=True)
|
||||
```
|
||||
|
||||
**列重命名**: 行业因子列名前缀为 `industry_`
|
||||
|
||||
---
|
||||
|
||||
## 附录：符号说明
|
||||
|
||||
| 符号 | 含义 |
|
||||
|------|------|
|
||||
| `epsilon` | 极小值 (1e-10)，防止除零 |
|
||||
| `mean(x, N)` | N周期滚动平均值 |
|
||||
| `std(x, N)` | N周期滚动标准差 |
|
||||
| `var(x, N)` | N周期滚动方差 |
|
||||
| `sum(x, N)` | N周期滚动求和 |
|
||||
| `max(x, N)` | N周期滚动最大值 |
|
||||
| `min(x, N)` | N周期滚动最小值 |
|
||||
| `diff(x, N)` | N周期差分 |
|
||||
| `pct_change(x, N)` | N周期百分比变化 |
|
||||
| `shift(x, N)` | N周期位移 |
|
||||
| `rank(x, pct=True)` | 截面排序 (百分比) |
|
||||
| `corr(x, y, N)` | N周期滚动相关系数 |
|
||||
| `cov(x, y, N)` | N周期滚动协方差 |
|
||||
| `skew(x, N)` | N周期滚动偏度 |
|
||||
| `kurt(x, N)` | N周期滚动峰度 |
|
||||
| `ATR` | Average True Range (TA-Lib) |
|
||||
| `OBV` | On-Balance Volume (TA-Lib) |
|
||||
| `RSI` | Relative Strength Index (TA-Lib) |
|
||||
| `SMA` | Simple Moving Average (TA-Lib) |
|
||||
| `EMA` | Exponential Moving Average (TA-Lib) |
|
||||
| `atan` | 反正切函数 |
|
||||
|
||||
---
|
||||
|
||||
*文档生成时间: 2026-03-06*
|
||||
*共收录 180+ 个因子*
|
||||
586
docs/factor_implementation_analysis.md
Normal file
586
docs/factor_implementation_analysis.md
Normal file
@@ -0,0 +1,586 @@
|
||||
# 因子表达式可实现性分析报告 (v2)
|
||||
|
||||
## 文档来源
|
||||
- **分析文件**: `docs/factor_expressions_document.md`
|
||||
- **分析日期**: 2026-03-06
|
||||
- **共收录因子**: 180+ 个
|
||||
- **更新说明**: 本版本明确区分 ts_* (时间序列) 和 cs_* (截面) 算子类型
|
||||
|
||||
---
|
||||
|
||||
## 一、当前 DSL 框架已实现功能
|
||||
|
||||
### 1.1 时间序列函数 (ts_*)
|
||||
|
||||
**说明**: 所有时间序列函数都按 `ts_code` 分组计算，防止跨股票数据泄露
|
||||
|
||||
| 函数名 | 参数 | 状态 | 说明 | Polars 实现 |
|
||||
|--------|------|------|------|-------------|
|
||||
| `ts_mean(x, window)` | x: 表达式, window: 窗口 | 已实现 | 滚动均值 | `rolling_mean(window).over("ts_code")` |
|
||||
| `ts_sum(x, window)` | x: 表达式, window: 窗口 | 已实现 | 滚动求和 | `rolling_sum(window).over("ts_code")` |
|
||||
| `ts_std(x, window)` | x: 表达式, window: 窗口 | 已实现 | 滚动标准差 | `rolling_std(window).over("ts_code")` |
|
||||
| `ts_max(x, window)` | x: 表达式, window: 窗口 | 已实现 | 滚动最大值 | `rolling_max(window).over("ts_code")` |
|
||||
| `ts_min(x, window)` | x: 表达式, window: 窗口 | 已实现 | 滚动最小值 | `rolling_min(window).over("ts_code")` |
|
||||
| `ts_delay(x, periods)` | x: 表达式, periods: 滞后 | 已实现 | 滞后 N 期 | `shift(periods).over("ts_code")` |
|
||||
| `ts_delta(x, periods)` | x: 表达式, periods: 差分 | 已实现 | 差分 N 期 | `(x - x.shift(periods)).over("ts_code")` |
|
||||
| `ts_corr(x, y, window)` | x, y: 表达式, window: 窗口 | 已实现 | 滚动相关系数 | `rolling_corr(y, window).over("ts_code")` |
|
||||
| `ts_cov(x, y, window)` | x, y: 表达式, window: 窗口 | 已实现 | 滚动协方差 | `rolling_cov(y, window).over("ts_code")` |
|
||||
| `ts_rank(x, window)` | x: 表达式, window: 窗口 | API 已定义 | 滚动排名 | 待实现 |
|
||||
|
||||
### 1.2 截面函数 (cs_*)
|
||||
|
||||
**说明**: 所有截面函数都按 `trade_date` 分组计算，防止跨日期数据泄露
|
||||
|
||||
| 函数名 | 参数 | 状态 | 说明 | Polars 实现 |
|
||||
|--------|------|------|------|-------------|
|
||||
| `cs_rank(x)` | x: 表达式 | 已实现 | 截面排名 (归一化到 [0,1]) | `(rank() / count()).over("trade_date")` |
|
||||
| `cs_zscore(x)` | x: 表达式 | 已实现 | Z-Score 标准化 | `((x - mean()) / std()).over("trade_date")` |
|
||||
| `cs_neutralize(x, group?)` | x: 表达式, group?: 分组列 | 已实现 | 行业/市值中性化 | `(x - mean()).over("trade_date")` |
|
||||
| `cs_winsorize(x, lower, upper)` | x: 表达式, lower/upper: 分位数 | API 已定义 | 缩尾处理 | 待实现 |
|
||||
| `cs_demean(x)` | x: 表达式 | API 已定义 | 去均值 | 待实现 |
|
||||
| `cs_std(x)` | x: 表达式 | 缺失 | 截面标准差 | 待实现 |
|
||||
| `cs_mean(x)` | x: 表达式 | 缺失 | 截面均值 | 待实现 |
|
||||
| `cs_sum(x)` | x: 表达式 | 缺失 | 截面求和 | 待实现 |
|
||||
| `cs_max(x)` | x: 表达式 | 缺失 | 截面最大值 | 待实现 |
|
||||
| `cs_min(x)` | x: 表达式 | 缺失 | 截面最小值 | 待实现 |
|
||||
|
||||
### 1.3 数学函数（无分组，逐元素计算）
|
||||
|
||||
| 函数名 | 参数 | 状态 | 说明 |
|
||||
|--------|------|------|------|
|
||||
| `log(x)` | x: 表达式 | API 已定义 | 自然对数 |
|
||||
| `log1p(x)` | x: 表达式 | 缺失 | log(1+x) |
|
||||
| `exp(x)` | x: 表达式 | API 已定义 | 指数函数 |
|
||||
| `sqrt(x)` | x: 表达式 | API 已定义 | 平方根 |
|
||||
| `sign(x)` | x: 表达式 | API 已定义 | 符号函数 (-1/0/1) |
|
||||
| `abs(x)` | x: 表达式 | API 已定义 | 绝对值 |
|
||||
| `max_(x, y)` | x, y: 表达式 | API 已定义 | 逐元素最大值 |
|
||||
| `min_(x, y)` | x, y: 表达式 | API 已定义 | 逐元素最小值 |
|
||||
| `clip(x, lower, upper)` | x: 表达式, lower/upper: 边界 | API 已定义 | 数值裁剪 |
|
||||
| `atan(x)` | x: 表达式 | 缺失 | 反正切函数 |
|
||||
|
||||
### 1.4 统计函数（需要新增）
|
||||
|
||||
| 函数名 | 类型 | 参数 | 说明 |
|
||||
|--------|------|------|------|
|
||||
| `ts_var(x, window)` | ts | x: 表达式, window: 窗口 | 滚动方差 |
|
||||
| `ts_skew(x, window)` | ts | x: 表达式, window: 窗口 | 滚动偏度 |
|
||||
| `ts_kurt(x, window)` | ts | x: 表达式, window: 窗口 | 滚动峰度 |
|
||||
| `ts_pct_change(x, periods)` | ts | x: 表达式, periods: 周期 | 百分比变化 |
|
||||
| `ts_ema(x, window)` | ts | x: 表达式, window: 窗口 | 指数移动平均 |
|
||||
| `ts_sma(x, window)` | ts | x: 表达式, window: 窗口 | 简单移动平均 (ts_mean 别名) |
|
||||
|
||||
### 1.5 TA-Lib 技术指标（用户可安装 TA-Lib）
|
||||
|
||||
| 函数名 | 类型 | 参数 | 说明 | 优先级 |
|
||||
|--------|------|------|------|--------|
|
||||
| `ts_atr(high, low, close, window)` | ts | high/low/close: 表达式, window: 窗口 | 真实波动幅度 | 高 |
|
||||
| `ts_obv(close, vol)` | ts | close/vol: 表达式 | 能量潮指标 | 高 |
|
||||
| `ts_rsi(close, window)` | ts | close: 表达式, window: 窗口 | 相对强弱指数 | 高 |
|
||||
| `ts_natr(high, low, close, window)` | ts | high/low/close: 表达式, window: 窗口 | 归一化 ATR | 中 |
|
||||
|
||||
**说明**: 由于用户已确认可安装 TA-Lib，建议直接使用 `ta-lib` Python 包，并通过 adapter 模式封装到 DSL 中。
|
||||
|
||||
### 1.6 条件函数
|
||||
|
||||
| 函数名 | 参数 | 状态 | 说明 |
|
||||
|--------|------|------|------|
|
||||
| `if_(condition, true_val, false_val)` | condition: 布尔表达式 | API 已定义 | 条件选择 |
|
||||
| `where(condition, true_val, false_val)` | 同上 | API 已定义 | if_ 的别名 |
|
||||
|
||||
### 1.7 符号/数据字段
|
||||
|
||||
| 字段名 | 状态 | 数据来源 | 所属类别 |
|
||||
|--------|------|----------|----------|
|
||||
| `close` | 可用 | daily/pro_bar 表 | 价格 |
|
||||
| `open` | 可用 | daily/pro_bar 表 | 价格 |
|
||||
| `high` | 可用 | daily/pro_bar 表 | 价格 |
|
||||
| `low` | 可用 | daily/pro_bar 表 | 价格 |
|
||||
| `vol` | 可用 | daily/pro_bar 表 | 成交量 |
|
||||
| `amount` | 可用 | daily/pro_bar 表 | 成交额 |
|
||||
| `pre_close` | 可用 | daily/pro_bar 表 | 价格 |
|
||||
| `change` | 可用 | daily/pro_bar 表 | 价格变化 |
|
||||
| `pct_chg` | 可用 | daily/pro_bar 表 | 涨跌幅 |
|
||||
| `turnover_rate` | 可用 | daily/pro_bar 表 | 换手率 |
|
||||
| `volume_ratio` | 可用 | daily/pro_bar 表 | 量比 |
|
||||
|
||||
### 1.8 支持的运算符
|
||||
|
||||
**算术运算符**: `+`, `-`, `*`, `/`, `**`, `//`, `%`
|
||||
**比较运算符**: `==`, `!=`, `<`, `<=`, `>`, `>=`
|
||||
**一元运算符**: `-`, `+`, `abs()`
|
||||
|
||||
---
|
||||
|
||||
## 二、需要新增的数据表
|
||||
|
||||
### 2.1 筹码分布数据（cyq 表）
|
||||
|
||||
**数据来源**: Tushare `cyq` API (VIP)
|
||||
**数据频率**: 日频
|
||||
**同步方式**: 按股票
|
||||
|
||||
| 字段名 | 类型 | 说明 | 依赖因子数 |
|
||||
|--------|------|------|------------|
|
||||
| `cost_5pct` | DOUBLE | 5% 成本价 | ~50个 |
|
||||
| `cost_15pct` | DOUBLE | 15% 成本价 | ~50个 |
|
||||
| `cost_50pct` | DOUBLE | 50% 成本价 (中位数) | ~50个 |
|
||||
| `cost_85pct` | DOUBLE | 85% 成本价 | ~50个 |
|
||||
| `cost_95pct` | DOUBLE | 95% 成本价 | ~50个 |
|
||||
| `weight_avg` | DOUBLE | 加权平均成本 | ~50个 |
|
||||
| `winner_rate` | DOUBLE | 获利盘比例 | ~50个 |
|
||||
| `his_high` | DOUBLE | 历史最高价 | ~10个 |
|
||||
| `his_low` | DOUBLE | 历史最低价 | ~10个 |
|
||||
|
||||
**依赖筹码数据的主要因子**:
|
||||
- `cashflow_to_ev_factor`
|
||||
- `chip_concentration_range`
|
||||
- `profit_pressure`
|
||||
- `underwater_resistance`
|
||||
- `vol_amp_loss_20`
|
||||
- `smallcap_concentration`
|
||||
- `cat_golden_resonance`
|
||||
- ...等 50+ 个
|
||||
|
||||
### 2.2 资金流向数据（moneyflow 表）
|
||||
|
||||
**数据来源**: Tushare `moneyflow` API (VIP)
|
||||
**数据频率**: 日频
|
||||
**同步方式**: 按股票
|
||||
|
||||
| 字段名 | 类型 | 说明 | 依赖因子数 |
|
||||
|--------|------|------|------------|
|
||||
| `buy_sm_vol` | DOUBLE | 小单买入量 | ~30个 |
|
||||
| `sell_sm_vol` | DOUBLE | 小单卖出量 | ~30个 |
|
||||
| `buy_lg_vol` | DOUBLE | 大单买入量 | ~30个 |
|
||||
| `sell_lg_vol` | DOUBLE | 大单卖出量 | ~30个 |
|
||||
| `buy_elg_vol` | DOUBLE | 特大单买入量 | ~30个 |
|
||||
| `sell_elg_vol` | DOUBLE | 特大单卖出量 | ~30个 |
|
||||
| `net_mf_vol` | DOUBLE | 净流入量 | ~20个 |
|
||||
|
||||
**依赖资金流向数据的主要因子**:
|
||||
- `lg_elg_net_buy_vol`
|
||||
- `flow_lg_elg_intensity`
|
||||
- `sm_net_buy_vol`
|
||||
- `flow_divergence_ratio`
|
||||
- `lg_flow_mom_corr_20_60`
|
||||
- `cs_rank_net_lg_flow_val`
|
||||
- ...等 30+ 个
|
||||
|
||||
### 2.3 财务数据（financial 表）
|
||||
|
||||
**数据来源**: Tushare `income`/`balance_sheet`/`cashflow` API (VIP)
|
||||
**数据频率**: 季度
|
||||
**同步方式**: 按股票，需 `asof_backward` join
|
||||
|
||||
| 字段名 | 类型 | 说明 | 数据来源表 |
|
||||
|--------|------|------|------------|
|
||||
| `bps` | DOUBLE | 每股净资产 | income |
|
||||
| `total_liab` | DOUBLE | 总负债 | balance_sheet |
|
||||
| `money_cap` | DOUBLE | 货币资金 | balance_sheet |
|
||||
| `n_cashflow_act` | DOUBLE | 经营活动现金流净额 | cashflow |
|
||||
| `undist_profit_ps` | DOUBLE | 每股未分配利润 | income |
|
||||
| `ocfps` | DOUBLE | 每股经营现金流 | income |
|
||||
| `roa` | DOUBLE | 资产回报率 | income |
|
||||
| `roe` | DOUBLE | 净资产收益率 | income |
|
||||
| `total_mv` | DOUBLE | 总市值 | daily_basic |
|
||||
| `circ_mv` | DOUBLE | 流通市值 | daily_basic |
|
||||
|
||||
**依赖财务数据的主要因子**:
|
||||
- `undist_profit_ps`
|
||||
- `cashflow_to_ev_factor`
|
||||
- `book_to_price_ratio`
|
||||
- `cs_rank_size`
|
||||
|
||||
### 2.4 行业分类数据
|
||||
|
||||
| 字段名 | 类型 | 说明 | 数据来源 |
|
||||
|--------|------|------|----------|
|
||||
| `cat_l2_code` | VARCHAR | 二级行业分类代码 | stock_basic |
|
||||
|
||||
**用途**: `daily_industry_deviation`, `cs_rank_ind_adj_lg_flow` 等分组计算
|
||||
|
||||
---
|
||||
|
||||
## 三、因子表达式可实现性分类
|
||||
|
||||
### 3.1 完全可实现（使用现有功能）
|
||||
|
||||
#### 技术指标因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `turnover_rate_mean_5` | `ts_mean(turnover_rate, 5)` | 低 |
|
||||
| `bbi_ratio_factor` | `(ts_mean(close,3)+ts_mean(close,6)+ts_mean(close,12)+ts_mean(close,24))/4/close` | 低 |
|
||||
|
||||
#### ARBR 因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `AR` | `ts_sum(high - open, N) / ts_sum(open - low, N) * 100` | 低 |
|
||||
| `BR` | `ts_sum(max_(0, high - ts_delay(close,1)), N) / ts_sum(max_(0, ts_delay(close,1) - low), N) * 100` | 中 |
|
||||
| `AR_BR` | `AR - BR` | 低 |
|
||||
|
||||
#### 成交量因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `volume_change_rate` | `ts_mean(vol, 2) / ts_mean(vol, 10) - 1` | 低 |
|
||||
| `turnover_deviation` | `(turnover_rate - ts_mean(turnover_rate, 3)) / ts_std(turnover_rate, 3)` | 低 |
|
||||
| `vol_std_5` | `ts_std(ts_delta(vol), 5)` | 低 |
|
||||
|
||||
#### 收益率因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `return_5` | `close / ts_delay(close, 5) - 1` | 低 |
|
||||
| `return_20` | `close / ts_delay(close, 20) - 1` | 低 |
|
||||
| `std_return_5` | `ts_std(ts_delta(close)/ts_delay(close), 5)` | 中 |
|
||||
| `std_return_90` | `ts_std(ts_delta(close)/ts_delay(close), 90)` | 中 |
|
||||
|
||||
#### 截面排序因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `cs_rank_volume_ratio` | `cs_rank(volume_ratio)` | 低 |
|
||||
| `cs_rank_turnover_rate` | `cs_rank(turnover_rate)` | 低 |
|
||||
|
||||
### 3.2 需要新增 ts_* 函数
|
||||
|
||||
| 因子名称 | 所需函数 | 表达式示例 | 优先级 |
|
||||
|----------|----------|------------|--------|
|
||||
| `variance_20` | `ts_var(x, window)` | `ts_var(pct_chg, 20)` | 高 |
|
||||
| `return_skew` | `ts_skew(x, window)` | `ts_skew(pct_chg, 5)` | 高 |
|
||||
| `return_kurtosis` | `ts_kurt(x, window)` | `ts_kurt(pct_chg, 5)` | 高 |
|
||||
| `turnover_diff_skew_20` | `ts_skew(x, window)` | `ts_skew(ts_delta(turnover_rate), 20)` | 高 |
|
||||
| `act_factor1` | `ts_ema(x, window)` | `atan((ts_ema(close,5)/ts_delay(ts_ema(close,5),1)-1)*100) * 57.3 / 50` | 高 |
|
||||
| `act_factor2` | `ts_ema(x, window)` | `atan((ts_ema(close,13)/ts_delay(ts_ema(close,13),1)-1)*100) * 57.3 / 40` | 高 |
|
||||
| `act_factor3` | `ts_ema(x, window)` | `atan((ts_ema(close,20)/ts_delay(ts_ema(close,20),1)-1)*100) * 57.3 / 21` | 高 |
|
||||
| `act_factor4` | `ts_ema(x, window)` | `atan((ts_ema(close,60)/ts_delay(ts_ema(close,60),1)-1)*100) * 57.3 / 10` | 高 |
|
||||
| `atr_14` | `ts_atr(h,l,c,w)` | `ts_atr(high, low, close, 14)` | 高 |
|
||||
| `atr_6` | `ts_atr(h,l,c,w)` | `ts_atr(high, low, close, 6)` | 高 |
|
||||
| `rsi_3` | `ts_rsi(x, window)` | `ts_rsi(close, 3)` | 高 |
|
||||
| `obv` | `ts_obv(c, v)` | `ts_obv(close, vol)` | 中 |
|
||||
| `maobv_6` | `ts_obv` + `ts_mean` | `ts_mean(ts_obv(close, vol), 6)` | 中 |
|
||||
|
||||
### 3.3 需要新增 cs_* 函数
|
||||
|
||||
| 因子名称 | 所需函数 | 表达式示例 | 优先级 |
|
||||
|----------|----------|------------|--------|
|
||||
| `daily_positive_benchmark` | `cs_mean(x)` | `cs_mean(pct_chg[pct_chg > 0])` | 高 |
|
||||
| `daily_industry_positive_benchmark` | `cs_mean(x)` by group | `cs_mean(pct_chg) by cat_l2_code` | 高 |
|
||||
| `industry_avg_flow` | `cs_mean(x)` by group | `cs_mean(net_lg_flow_vol) by trade_date, cat_l2_code` | 中 |
|
||||
|
||||
**注意**: 这些因子需要 **分组聚合** 能力，可能需要扩展 DSL 语法支持 `by` 子句，或通过 `cs_neutralize` 的 group 参数实现。
|
||||
|
||||
### 3.4 需要新增数据表
|
||||
|
||||
#### 依赖筹码数据（cyq 表）的因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `chip_concentration_range` | `(cost_95pct - cost_5pct) / (close + epsilon)` | 低 |
|
||||
| `chip_skewness` | `(weight_avg - cost_50pct) / (cost_50pct + epsilon)` | 低 |
|
||||
| `profit_pressure` | `winner_rate * 0.5 * ((close/cost_85pct-1) + (close/cost_95pct-1))` | 中 |
|
||||
| `underwater_resistance` | `(1-winner_rate) * max_(0, cost_15pct-close) / (close+epsilon)` | 中 |
|
||||
| `smallcap_concentration` | `(1 / log(circ_mv)) * (cost_85pct - cost_15pct)` | 中 |
|
||||
|
||||
#### 依赖资金流向数据（moneyflow 表）的因子
|
||||
| 因子名称 | 表达式 | 复杂度 |
|
||||
|----------|--------|--------|
|
||||
| `lg_elg_net_buy_vol` | `(buy_lg_vol + buy_elg_vol - sell_lg_vol - sell_elg_vol)` | 低 |
|
||||
| `flow_lg_elg_intensity` | `lg_elg_net_buy_vol / (vol + epsilon)` | 低 |
|
||||
| `cs_rank_net_lg_flow_val` | `cs_rank(lg_elg_net_buy_vol * close)` | 低 |
|
||||
|
||||
#### 依赖财务数据的因子
|
||||
| 因子名称 | 表达式 | 复杂度 | 特殊需求 |
|
||||
|----------|--------|--------|----------|
|
||||
| `book_to_price_ratio` | `bps / close` | 低 | 需 asof_backward join |
|
||||
| `cashflow_to_ev_factor` | `n_cashflow_act / (total_mv*10000 + total_liab - money_cap)` | 高 | 需多表 join |
|
||||
|
||||
### 3.5 需要复杂条件逻辑
|
||||
|
||||
| 因子名称 | 逻辑 | 实现方式 |
|
||||
|----------|------|----------|
|
||||
| `daily_deviation` | 按条件选择不同基准 | `if_(pct_chg > 0, pct_chg - pos_benchmark, if_(pct_chg < 0, pct_chg - neg_benchmark, 0))` |
|
||||
| `cat_volume_breakout` | 布尔分类 | `if_(vol > ts_max(vol, 5), 1, 0)` |
|
||||
| `cat_turnover_spike` | 阈值分类 | `if_(turnover_rate > ts_mean(turnover_rate,3) + 2*ts_std(turnover_rate,3), 1, 0)` |
|
||||
| `cat_golden_resonance` | 多条件组合 | `(close > weight_avg) & (volume_ratio > 1.5) & (winner_rate > 0.7)` |
|
||||
|
||||
---
|
||||
|
||||
## 四、实施计划（按优先级排序）
|
||||
|
||||
### Phase 1: 核心 ts_* 函数（Week 1-2）
|
||||
|
||||
**目标**: 实现最常用的 30+ 个因子
|
||||
|
||||
```python
|
||||
# 1. 新增数学函数（逐元素，无分组）
|
||||
def atan(x): ...
|
||||
def log1p(x): ...
|
||||
|
||||
# 2. 新增统计函数（ts_*，按 ts_code 分组）
|
||||
def ts_var(x, window): ... # 滚动方差
|
||||
def ts_skew(x, window): ... # 滚动偏度
|
||||
def ts_kurt(x, window): ... # 滚动峰度
|
||||
def ts_pct_change(x, periods): ... # 百分比变化
|
||||
|
||||
# 3. 新增 EMA（ts_*，使用 Polars ewm_mean）
|
||||
def ts_ema(x, window):
|
||||
# 使用 Polars: pl.col(x).ewm_mean(span=window).over("ts_code")
|
||||
pass
|
||||
|
||||
# 4. 实现 ATR（基于现有函数组合）
|
||||
def ts_atr(high, low, close, window):
|
||||
tr1 = high - low
|
||||
tr2 = abs(high - ts_delay(close, 1))
|
||||
tr3 = abs(low - ts_delay(close, 1))
|
||||
tr = max_(tr1, max_(tr2, tr3))
|
||||
return ts_mean(tr, window)
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `variance_20`
|
||||
- `return_skew`, `return_kurtosis`
|
||||
- `turnover_diff_skew_20`
|
||||
- `act_factor1-4`
|
||||
- `atr_14`, `atr_6`
|
||||
- `vol_adj_roc_20`
|
||||
|
||||
### Phase 2: TA-Lib 集成（Week 3）
|
||||
|
||||
**目标**: 支持 RSI、OBV 等复杂技术指标
|
||||
|
||||
```python
|
||||
# 安装 TA-Lib
|
||||
# pip install TA-Lib
|
||||
|
||||
# 封装为 ts_* 函数
|
||||
def ts_rsi(close, window):
|
||||
# 使用 talib.RSI，并添加 .over("ts_code")
|
||||
pass
|
||||
|
||||
def ts_obv(close, vol):
|
||||
# 使用 talib.OBV，并添加 .over("ts_code")
|
||||
pass
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `rsi_3`
|
||||
- `obv`
|
||||
- `maobv_6`
|
||||
|
||||
### Phase 3: 筹码分布数据同步（Week 4）
|
||||
|
||||
**目标**: 实现 cyq 表同步，支持 50+ 个筹码相关因子
|
||||
|
||||
```python
|
||||
class CyqSync(StockBasedSync):
|
||||
table_name = "cyq"
|
||||
TABLE_SCHEMA = {
|
||||
"ts_code": "VARCHAR(16) NOT NULL",
|
||||
"trade_date": "DATE NOT NULL",
|
||||
"cost_5pct": "DOUBLE",
|
||||
"cost_15pct": "DOUBLE",
|
||||
"cost_50pct": "DOUBLE",
|
||||
"cost_85pct": "DOUBLE",
|
||||
"cost_95pct": "DOUBLE",
|
||||
"weight_avg": "DOUBLE",
|
||||
"winner_rate": "DOUBLE",
|
||||
"his_high": "DOUBLE",
|
||||
"his_low": "DOUBLE",
|
||||
}
|
||||
PRIMARY_KEY = ("ts_code", "trade_date")
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `chip_concentration_range`
|
||||
- `profit_pressure`
|
||||
- `underwater_resistance`
|
||||
- `smallcap_concentration`
|
||||
- ...等 50+ 个
|
||||
|
||||
### Phase 4: 资金流向数据同步（Week 5）
|
||||
|
||||
**目标**: 实现 moneyflow 表同步，支持 30+ 个资金相关因子
|
||||
|
||||
```python
|
||||
class MoneyFlowSync(StockBasedSync):
|
||||
table_name = "moneyflow"
|
||||
TABLE_SCHEMA = {
|
||||
"ts_code": "VARCHAR(16) NOT NULL",
|
||||
"trade_date": "DATE NOT NULL",
|
||||
"buy_sm_vol": "DOUBLE",
|
||||
"sell_sm_vol": "DOUBLE",
|
||||
"buy_lg_vol": "DOUBLE",
|
||||
"sell_lg_vol": "DOUBLE",
|
||||
"buy_elg_vol": "DOUBLE",
|
||||
"sell_elg_vol": "DOUBLE",
|
||||
"net_mf_vol": "DOUBLE",
|
||||
}
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `lg_elg_net_buy_vol`
|
||||
- `flow_lg_elg_intensity`
|
||||
- `cs_rank_net_lg_flow_val`
|
||||
- ...等 30+ 个
|
||||
|
||||
### Phase 5: 分组聚合功能（Week 6）
|
||||
|
||||
**目标**: 实现 cs_mean by group 等分组计算
|
||||
|
||||
```python
|
||||
# 方式1: 扩展 cs_neutralize 的 group 参数
|
||||
def cs_neutralize(x, group=None):
|
||||
if group:
|
||||
return (x - x.mean().over(["trade_date", group]))
|
||||
return (x - x.mean().over("trade_date"))
|
||||
|
||||
# 方式2: 新增 cs_mean_by 函数
|
||||
def cs_mean_by(x, by):
|
||||
return x.mean().over(["trade_date"] + by)
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `daily_deviation`
|
||||
- `daily_industry_deviation`
|
||||
- `cs_rank_ind_adj_lg_flow`
|
||||
|
||||
### Phase 6: 财务数据集成（Week 7）
|
||||
|
||||
**目标**: 实现财务数据同步和 asof_backward join
|
||||
|
||||
```python
|
||||
# 使用 DataSpec 的 asof_backward 模式
|
||||
spec = DataSpec(
|
||||
table="financial_income",
|
||||
columns=["bps", "roe", "roa"],
|
||||
join_type="asof_backward",
|
||||
left_on="trade_date",
|
||||
right_on="f_ann_date",
|
||||
)
|
||||
```
|
||||
|
||||
**完成后可实现的因子**:
|
||||
- `book_to_price_ratio`
|
||||
- `cashflow_to_ev_factor`
|
||||
- `cs_rank_size` (需要 circ_mv)
|
||||
|
||||
---
|
||||
|
||||
## 五、可实现性统计
|
||||
|
||||
### 按实现难度分类
|
||||
|
||||
| 类别 | 数量 | 状态 | 时间预估 |
|
||||
|------|------|------|----------|
|
||||
| 立即可实现 | ~40个 | 使用现有功能 | 0 周 |
|
||||
| 需新增 ts_* 函数 | ~30个 | Phase 1-2 | 2-3 周 |
|
||||
| 需新增 cs_* 函数 | ~10个 | Phase 5 | 1 周 |
|
||||
| 需筹码数据 (cyq) | ~50个 | Phase 3 | 1 周 |
|
||||
| 需资金数据 (moneyflow) | ~30个 | Phase 4 | 1 周 |
|
||||
| 需财务数据 (financial) | ~20个 | Phase 6 | 1 周 |
|
||||
| **总计** | **180+个** | | **6-7 周** |
|
||||
|
||||
### 按数据源分类
|
||||
|
||||
| 数据源 | 依赖因子数 | 实现难度 | 优先级 |
|
||||
|--------|------------|----------|--------|
|
||||
| daily/pro_bar (已有) | ~40 | 低 | 高 |
|
||||
| 纯技术指标 (ts_*) | ~30 | 中 | 高 |
|
||||
| 筹码分布 (cyq) | ~50 | 中 | 中 |
|
||||
| 资金流向 (moneyflow) | ~30 | 中 | 中 |
|
||||
| 财务数据 (financial) | ~20 | 高 | 低 |
|
||||
| 分组计算 (cs_* by group) | ~10 | 高 | 低 |
|
||||
|
||||
---
|
||||
|
||||
## 六、立即可实现的因子示例
|
||||
|
||||
```python
|
||||
from src.factors.api import close, open, high, low, vol, pct_chg, turnover_rate
|
||||
from src.factors.api import ts_mean, ts_std, ts_max, ts_min, ts_delay, ts_delta, ts_corr, ts_sum
|
||||
from src.factors.api import cs_rank, cs_zscore, cs_neutralize
|
||||
from src.factors.api import log, sqrt, abs, sign, max_, min_, clip, if_
|
||||
|
||||
# 1. BBI 比率
|
||||
bbi = (ts_mean(close, 3) + ts_mean(close, 6) + ts_mean(close, 12) + ts_mean(close, 24)) / 4
|
||||
bbi_ratio_factor = bbi / close
|
||||
|
||||
# 2. AR 因子 (N=26)
|
||||
ar = ts_sum(high - open, 26) / ts_sum(open - low, 26) * 100
|
||||
|
||||
# 3. BR 因子 (N=26)
|
||||
prev_close = ts_delay(close, 1)
|
||||
br = ts_sum(max_(0, high - prev_close), 26) / ts_sum(max_(0, prev_close - low), 26) * 100
|
||||
ar_br = ar - br
|
||||
|
||||
# 4. 成交量变化率
|
||||
volume_change_rate = ts_mean(vol, 2) / ts_mean(vol, 10) - 1
|
||||
|
||||
# 5. 换手率偏离度
|
||||
turnover_deviation = (turnover_rate - ts_mean(turnover_rate, 3)) / ts_std(turnover_rate, 3)
|
||||
|
||||
# 6. 5日收益率
|
||||
return_5 = close / ts_delay(close, 5) - 1
|
||||
|
||||
# 7. 20日收益率
|
||||
return_20 = close / ts_delay(close, 20) - 1
|
||||
|
||||
# 8. Alpha 003
|
||||
alpha_003 = if_(high != low, (close - open) / (high - low), 0)
|
||||
|
||||
# 9. Alpha 007 (等 Phase 1 后)
|
||||
# alpha_007 = cs_rank(ts_corr(close, vol, 5))
|
||||
|
||||
# 10. 截面排名
|
||||
rank_close = cs_rank(close)
|
||||
rank_volume = cs_rank(vol)
|
||||
|
||||
# 11. Z-Score 标准化
|
||||
zscore_close = cs_zscore(close)
|
||||
|
||||
# 12. 截面中性化 (Phase 5 后支持分组)
|
||||
# neutralized = cs_neutralize(close, group="industry")
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 七、结论与建议
|
||||
|
||||
### 关键发现
|
||||
|
||||
1. **ts_* vs cs_* 区分清晰**:
|
||||
- `ts_*` 函数按 `ts_code` 分组，用于时间序列计算
|
||||
- `cs_*` 函数按 `trade_date` 分组，用于截面计算
|
||||
- 数学函数（log, exp, atan 等）为逐元素计算，无分组
|
||||
|
||||
2. **当前框架已实现**:
|
||||
- 9个核心 ts_* 函数（mean, std, max, min, sum, delay, delta, corr, cov）
|
||||
- 3个 cs_* 函数（rank, zscore, neutralize）
|
||||
- 完整的运算符重载系统
|
||||
|
||||
3. **主要缺口**:
|
||||
- 统计函数: `ts_var`, `ts_skew`, `ts_kurt`, `ts_pct_change`
|
||||
- 技术指标: `ts_ema`, `ts_atr`, `ts_rsi`, `ts_obv`
|
||||
- 数学函数: `atan`, `log1p`
|
||||
- 分组聚合: `cs_mean`, `cs_sum` by group
|
||||
- 数据表: cyq, moneyflow, financial
|
||||
|
||||
### 推荐实施路径
|
||||
|
||||
| 阶段 | 目标 | 时间 | 完成后可计算因子数 |
|
||||
|------|------|------|-------------------|
|
||||
| Phase 1 | ts_var, ts_skew, ts_kurt, ts_ema, atan | 1-2周 | +30个 |
|
||||
| Phase 2 | TA-Lib 集成 (RSI, OBV) | 1周 | +5个 |
|
||||
| Phase 3 | 筹码数据同步 (cyq) | 1周 | +50个 |
|
||||
| Phase 4 | 资金流向同步 (moneyflow) | 1周 | +30个 |
|
||||
| Phase 5 | 分组聚合功能 | 1周 | +10个 |
|
||||
| Phase 6 | 财务数据集成 | 1周 | +20个 |
|
||||
| **总计** | | **6-7周** | **180+个** |
|
||||
|
||||
### 建议优先实现的因子（高 ROI）
|
||||
|
||||
1. **BBI 比率**: 纯技术指标，无需新数据
|
||||
2. **AR/BR**: 经典情绪指标，无需新数据
|
||||
3. **换手率偏离度**: 流动性因子，无需新数据
|
||||
4. **筹码集中度**: 需 cyq 数据，但实现简单
|
||||
5. **大单净流入**: 需 moneyflow 数据，实现简单
|
||||
|
||||
通过以上步骤，预计在 **6-7 周** 内可实现文档中 **90% 以上的因子表达式**。
|
||||
Reference in New Issue
Block a user