# 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*