Contents

中文EN

Theory: Factor Models & Styles

1. Single-Factor ModelsCode2. Multi-Factor Models3. Classic Style Factors4. Expanding the Factor Universe

Practice: Factor Production & Management

5. Factor Construction and Neutralization6. Factor Testing FrameworkCode7. Portfolio Optimization and Attribution8. Factor Timing and Crowding

Extension: Macro Quant & Asset Allocation

9. Macro Data and Factors10. Macro Regime Identification11. Asset AllocationCode12. A Decade of Macro Quant NotesCode

12. A Decade of Macro Quant Notes

Over roughly the past decade, the macro forces most often cited behind A-share and bond moves remain growth, inflation, and rates (plus credit when needed). This chapter builds a reproducible ten-year note.

1. Construct the classic macro factors above from public data (monthly);
2. Regress CSI 300 monthly returns on those factors to estimate loadings;
3. Aggregate factor contributions $= \beta_f \times f_t$ by calendar year to answer: in each year, which macro factor explained / contributed the most to equity returns?

12.1 Analytical Framework

Factor definitions (growth, inflation, rates):

Factor	Proxy	Monthly factor value
Growth	Official manufacturing PMI	$\Delta$ PMI
Inflation	CPI YoY	$\Delta$ CPI YoY (pp)
Rates	10Y CGB yield	$\Delta$ yield (pp); rising yields usually compress valuations
Credit	Aggregate financing stock YoY (if available)	$\Delta$ financing YoY

Contribution decomposition:

For each calendar year $y$, run a monthly regression on that year (or a rolling window)

$$ r_t = \alpha_y + \sum_f \beta_{y,f} F_{f,t-1} + \varepsilon_t $$

where $F$ are standardized factors (for comparability), and a one-period lag reduces contemporaneous reverse causality. Annual contribution is

$$ C_{y,f} = \sum_{t \in y} \hat\beta_{y,f} F_{f,t-1}. $$

Rank by $|C_{y,f}|$ (sign shows direction: positive = co-moving lift to equities; negative = drag).

12.2 Data Preparation

from pathlib import Path
import warnings
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt

warnings.filterwarnings("ignore")
plt.rcParams["axes.unicode_minus"] = False
try:
    plt.rcParams["font.sans-serif"] = ["Microsoft YaHei", "SimHei", "DejaVu Sans"]
except Exception:
    pass

DATA_DIR = Path("Data/macro_decade")
DATA_DIR.mkdir(parents=True, exist_ok=True)
CACHE = DATA_DIR / "macro_factors_monthly.parquet"

def _month_end(s: pd.Series) -> pd.Series:
    s = s.copy
    s.index = pd.to_datetime(s.index)
    return s.sort_index.resample("ME").last

def download_or_load -> pd.DataFrame:
    if CACHE.exists:
        df = pd.read_parquet(CACHE)
        df.index = pd.to_datetime(df.index)
        print(f"loaded cache: {CACHE}  rows={len(df)}")
        return df

    import akshare as ak

    # PMI
    pmi = ak.macro_china_pmi.rename(columns=lambda x: str(x).strip)
    dcol = [c for c in pmi.columns if "月" in c or "日期" in c][0]
    vcol = [c for c in pmi.columns if "制造" in c or "指数" in c][0]
    raw = (
        pmi[dcol].astype(str)
        .str.replace("月份", "", regex=False)
        .str.replace("年", "-", regex=False)
        .str.replace("月", "", regex=False)
    )
    pmi[dcol] = pd.to_datetime(raw, errors="coerce")
    pmi_s = _month_end(pd.Series(pd.to_numeric(pmi[vcol], errors="coerce").values, index=pmi[dcol], name="pmi"))

    # CPI YoY
    cpi = ak.macro_china_cpi_yearly.rename(columns=lambda x: str(x).strip)
    dcol = [c for c in cpi.columns if "日期" in c or "月" in c][0]
    vcol = [c for c in cpi.columns if "今" in c or "同比" in c or "值" in c][-1]
    cpi[dcol] = pd.to_datetime(cpi[dcol], errors="coerce")
    cpi_s = _month_end(pd.Series(pd.to_numeric(cpi[vcol], errors="coerce").values, index=cpi[dcol], name="cpi_yoy"))

    # 10Y yield (CN–US bond yield API is more stable)
    y = ak.bond_zh_us_rate(start_date="20100101")
    y = y.rename(columns=lambda x: str(x).strip)
    dcol = "日期" if "日期" in y.columns else y.columns[0]
    tcol = "中国国债收益率10年" if "中国国债收益率10年" in y.columns else [c for c in y.columns if "10" in str(c) and "中国" in str(c)][0]
    y[dcol] = pd.to_datetime(y[dcol], errors="coerce")
    y10 = _month_end(pd.Series(pd.to_numeric(y[tcol], errors="coerce").values, index=y[dcol], name="y10"))
    print("y10 via bond_zh_us_rate:", tcol, "non-na=", y10.notna.sum)

    # social financing YoY (credit) — best effort
    credit_s = None
    for fn_name in ["macro_china_shrzgm", "macro_china_new_financial_credit"]:
        try:
            fn = getattr(ak, fn_name)
            sf = fn.rename(columns=lambda x: str(x).strip)
            # try find a yoy-like column; else skip
            dcol = [c for c in sf.columns if "日期" in c or "月" in c][0]
            sf[dcol] = pd.to_datetime(sf[dcol], errors="coerce")
            num_cols = [c for c in sf.columns if c != dcol]
            # prefer YoY column (同比 is Chinese API field name)
            yoy_cols = [c for c in num_cols if "同比" in str(c)]
            col = yoy_cols[0] if yoy_cols else num_cols[0]
            credit_s = _month_end(
                pd.Series(pd.to_numeric(sf[col], errors="coerce").values, index=sf[dcol], name="credit")
            )
            print(f"credit proxy via {fn_name}: {col}")
            break
        except Exception as e:
            print(f"skip credit source {fn_name}: {e}")

    # HS300
    hs = ak.stock_zh_index_daily(symbol="sh000300")
    hs["date"] = pd.to_datetime(hs["date"])
    hs = hs.set_index("date").sort_index
    px = hs["close"] if "close" in hs.columns else hs.iloc[:, 0]
    stock = _month_end(px).pct_change.rename("stock_ret")

    parts = [pmi_s, cpi_s, y10, stock]
    if credit_s is not None:
        parts.append(credit_s)
    df = pd.concat(parts, axis=1).sort_index
    df = df.loc["2014-01-01":].copy
    df.to_parquet(CACHE)
    print(f"saved {CACHE} rows={len(df)}")
    return df

raw = download_or_load
print(raw.tail)
print("coverage:", raw.index.min.date, "→", raw.index.max.date)

  0%|          | 0/19 [00:00<?, ?it/s]

y10 via bond_zh_us_rate: 中国国债收益率10年 non-na= 200
skip credit source macro_china_shrzgm: 'NaTType' object has no attribute 'normalize'
skip credit source macro_china_new_financial_credit: 'NaTType' object has no attribute 'normalize'
saved Data\macro_decade\macro_factors_monthly.parquet rows=152
             pmi  cpi_yoy     y10  stock_ret
2026-04-30  50.3      NaN  1.7473   0.080282
2026-05-31  50.0      NaN  1.7090   0.017643
2026-06-30  50.3      NaN  1.7330   0.017847
2026-07-31  49.2      NaN  1.7141  -0.078570
2026-08-31   NaN      NaN  1.7114   0.023155
coverage: 2014-01-31 → 2026-08-31

12.3 Build Factors and Annual Contribution Decomposition

Growth and inflation use first differences; rates use changes in the 10Y yield. Apply a full-sample z-score so $\beta\times F$ magnitudes are comparable. Factors enter the regression lagged one month.

# --- build factors ---
fac = pd.DataFrame(index=raw.index)
fac["Growth"] = raw["pmi"].diff
fac["Inflation"] = raw["cpi_yoy"].diff
fac["Rates"] = raw["y10"].diff
if "credit" in raw.columns and raw["credit"].notna.sum > 36:
    # if level looks like YoY already, take diff; if huge levels, still diff
    fac["Credit"] = raw["credit"].diff
    print("include Credit factor")
else:
    print("Credit factor data insufficient; this note uses Growth / Inflation / Rates")

fac = fac.dropna(how="all")
# standardize
fac_z = (fac - fac.mean) / fac.std(ddof=0)
stock = raw["stock_ret"]

factor_cols = list(fac_z.columns)
panel = fac_z.shift(1).join(stock, how="inner").dropna(subset=factor_cols + ["stock_ret"])
panel["year"] = panel.index.year
# focus on last ~10 calendar years with enough months
years = sorted([y for y in panel["year"].unique if y >= 2016])
panel = panel[panel["year"].isin(years)].copy
factors = [c for c in factor_cols if c in panel.columns]
print("factors:", factors)
print("sample months:", len(panel), "years:", years[0], "→", years[-1])
panel.tail

信用因子数据不足，本笔记使用增长/通胀/利率三因子
factors: ['增长', '通胀', '利率']
sample months: 118 years: 2016 → 2025

	增长	通胀	利率	stock_ret	year
2025-06-30	0.266108	1.286695	0.550046	0.024959	2025
2025-07-31	0.109163	0.044369	-0.048963	0.035444	2025
2025-08-31	-0.204725	0.044369	0.639224	0.103339	2025
2025-09-30	0.056849	0.458478	1.278616	0.032009	2025
2025-10-31	0.213793	-0.162686	0.345609	-0.000006	2025

from numpy.linalg import lstsq

MIN_OBS = 8  # months in a year to estimate

def year_regression(df_y: pd.DataFrame, factors: list[str]):
    """OLS stock_ret on factors; return betas, contrib series, R2."""
    y = df_y["stock_ret"].values.astype(float)
    X = np.column_stack([np.ones(len(df_y))] + [df_y[f].values.astype(float) for f in factors])
    coef, *_ = lstsq(X, y, rcond=None)
    alpha = coef[0]
    betas = dict(zip(factors, coef[1:]))
    fitted = X @ coef
    ss_res = ((y - fitted) ** 2).sum
    ss_tot = ((y - y.mean) ** 2).sum
    r2 = 1 - ss_res / ss_tot if ss_tot > 0 else np.nan
    contrib = {f: betas[f] * df_y[f] for f in factors}
    return alpha, betas, contrib, r2

rows = []
contrib_yearly = []
beta_yearly = []

for y in years:
    df_y = panel.loc[panel["year"] == y, factors + ["stock_ret"]].dropna
    if len(df_y) < MIN_OBS:
        print(f"skip {y}: only {len(df_y)} months")
        continue
    alpha, betas, contrib, r2 = year_regression(df_y, factors)
    stock_ann = (1 + df_y["stock_ret"]).prod - 1
    c_sum = {f: float(contrib[f].sum) for f in factors}
    # winner by absolute contribution
    winner = max(c_sum, key=lambda k: abs(c_sum[k]))
    rows.append({
        "Year": y,
        "Stock ann. return": stock_ann,
        "R2": r2,
        "Dominant factor": winner,
        "Dominant contrib.": c_sum[winner],
        **{f"Contrib_{f}": c_sum[f] for f in factors},
        **{f"beta_{f}": betas[f] for f in factors},
    })
    for f in factors:
        contrib_yearly.append({"Year": y, "Factor": f, "Contribution": c_sum[f]})
        beta_yearly.append({"Year": y, "Factor": f, "beta": betas[f]})

summary = pd.DataFrame(rows).set_index("Year")
contrib_df = pd.DataFrame(contrib_yearly)
beta_df = pd.DataFrame(beta_yearly)

print("[Dominant macro factor by year]")
show_cols = ["Stock ann. return", "R2", "Dominant factor", "Dominant contrib."] + [f"Contrib_{f}" for f in factors]
print(summary[show_cols].round(4).to_string)

summary.to_csv(DATA_DIR / "yearly_factor_contribution.csv", encoding="utf-8-sig")
contrib_df.to_csv(DATA_DIR / "contrib_long.csv", encoding="utf-8-sig")
print("saved tables →", DATA_DIR.resolve)

【逐年主导宏观因子】
       股票年收益      R2 主导因子    主导贡献   贡献_增长   贡献_通胀   贡献_利率
年份                                                       
2016 -0.1128  0.2610   增长 -0.1100 -0.1100 -0.0610  0.0608
2017  0.2178  0.1485   利率  0.0740  0.0044  0.0001  0.0740
2018 -0.2531  0.1990   通胀  0.0467 -0.0085  0.0467 -0.0042
2019  0.3607  0.5440   通胀 -0.0469 -0.0069 -0.0469 -0.0105
2020  0.2721  0.7104   通胀  0.0639  0.0122  0.0639 -0.0042
2021 -0.0520  0.5050   利率  0.0305  0.0120  0.0247  0.0305
2022 -0.2163  0.1281   利率 -0.0550  0.0188  0.0220 -0.0550
2023 -0.1138  0.1653   通胀  0.0383  0.0098  0.0383 -0.0011
2024  0.1468  0.1008   利率 -0.1801  0.0025 -0.0198 -0.1801
2025  0.1794  0.3271   增长  0.0118  0.0118 -0.0004  0.0039
saved tables → D:\A_Topics\202607_03_TidyFinanceAShare\202608_02_传统量化\《因子投资进阶：从单因子到多因子体系》_notebooks\Data\macro_decade

12.4 Visualization: Who Contributed Most Each Year?

# stacked bar of signed contributions + winner annotation
pivot = contrib_df.pivot(index="Year", columns="Factor", values="Contribution").reindex(columns=factors)

fig, axes = plt.subplots(2, 1, figsize=(11, 7.5), gridspec_kw={"height_ratios": [2.2, 1.2]})

# 1) contribution bars
ax = axes[0]
x = np.arange(len(pivot.index))
bottom_pos = np.zeros(len(pivot))
bottom_neg = np.zeros(len(pivot))
colors = {"Growth": "#4C78A8", "Inflation": "#F58518", "Rates": "#54A24B", "Credit": "#E45756"}
for f in factors:
    vals = pivot[f].fillna(0).values
    pos = np.where(vals >= 0, vals, 0.0)
    neg = np.where(vals < 0, vals, 0.0)
    ax.bar(x, pos, bottom=bottom_pos, label=f, color=colors.get(f, None), width=0.75)
    ax.bar(x, neg, bottom=bottom_neg, color=colors.get(f, None), width=0.75)
    bottom_pos += pos
    bottom_neg += neg
ax.axhline(0, color="#333", lw=0.8)
ax.set_xticks(x)
ax.set_xticklabels(pivot.index.astype(int))
ax.set_ylabel("Annual contribution Σ β·F (standardized factors)")
ax.set_title("CSI 300: annual contribution of classic macro factors")
ax.legend(ncol=len(factors), frameon=False, loc="upper right")

# annotate winner
for i, y in enumerate(pivot.index):
    w = summary.loc[y, "Dominant factor"] if y in summary.index else ""
    v = summary.loc[y, "Dominant contrib."] if y in summary.index else 0
    ymin, ymax = ax.get_ylim
    ax.text(i, ymax * 0.92, f"{w}", ha="center", va="top", fontsize=9, color="#333")

# 2) |contribution| share
ax2 = axes[1]
abs_share = pivot.abs.div(pivot.abs.sum(axis=1), axis=0)
abs_share.plot(kind="bar", stacked=True, ax=ax2, color=[colors.get(f, None) for f in factors], width=0.75, legend=False)
ax2.set_ylabel("|Contribution| share")
ax2.set_xlabel("")
ax2.set_title("Share of absolute contribution by factor each year")
ax2.set_xticklabels(abs_share.index.astype(int), rotation=0)

plt.tight_layout
fig_path = DATA_DIR / "decade_factor_contribution.png"
plt.savefig(fig_path, dpi=140, bbox_inches="tight")
plt.show
plt.close
print("figure:", fig_path)

figure: Data\macro_decade\decade_factor_contribution.png

12.5 Reading the Chart (Note-Style Takeaways)

The next cell auto-generates a one-line summary per year for research notes or resume project write-ups. When interpreting:

• Growth: PMI improvement usually maps to risk appetite and earnings expectations;
• Inflation: higher prints can lift nominal growth expectations or trigger tightening fears;
• Rates: rising yields often compress equity valuations (contributions are often negative);
• If a year's $R^2$ is very low, microstructure / policy / liquidity residuals dominate and the three macro factors are not enough.

lines = []
for y, row in summary.iterrows:
    parts = ", ".join(f"{f}={row[f'Contrib_{f}']:+.3f}" for f in factors)
    lines.append(
        f"{int(y)}: stock return {row['Stock ann. return']:+.1%}, "
        f"macro regression R²={row['R2']:.2f}; "
        f"largest |contrib.| = [{row['Dominant factor']}] ({row['Dominant contrib.']:+.3f}). "
        f"Breakdown: {parts}."
    )

note = "\n".join(lines)
print(note)
(DATA_DIR / "decade_note.txt").write_text(note, encoding="utf-8")

# frequency of winners
print("\n[Dominant-factor counts over the decade]")
print(summary["Dominant factor"].value_counts.to_string)

2016年：股票收益 -11.3%，宏观回归 R²=0.26；|贡献|最大为【增长】（-0.110）。分解：增长=-0.110, 通胀=-0.061, 利率=+0.061。
2017年：股票收益 +21.8%，宏观回归 R²=0.15；|贡献|最大为【利率】（+0.074）。分解：增长=+0.004, 通胀=+0.000, 利率=+0.074。
2018年：股票收益 -25.3%，宏观回归 R²=0.20；|贡献|最大为【通胀】（+0.047）。分解：增长=-0.009, 通胀=+0.047, 利率=-0.004。
2019年：股票收益 +36.1%，宏观回归 R²=0.54；|贡献|最大为【通胀】（-0.047）。分解：增长=-0.007, 通胀=-0.047, 利率=-0.010。
2020年：股票收益 +27.2%，宏观回归 R²=0.71；|贡献|最大为【通胀】（+0.064）。分解：增长=+0.012, 通胀=+0.064, 利率=-0.004。
2021年：股票收益 -5.2%，宏观回归 R²=0.51；|贡献|最大为【利率】（+0.030）。分解：增长=+0.012, 通胀=+0.025, 利率=+0.030。
2022年：股票收益 -21.6%，宏观回归 R²=0.13；|贡献|最大为【利率】（-0.055）。分解：增长=+0.019, 通胀=+0.022, 利率=-0.055。
2023年：股票收益 -11.4%，宏观回归 R²=0.17；|贡献|最大为【通胀】（+0.038）。分解：增长=+0.010, 通胀=+0.038, 利率=-0.001。
2024年：股票收益 +14.7%，宏观回归 R²=0.10；|贡献|最大为【利率】（-0.180）。分解：增长=+0.002, 通胀=-0.020, 利率=-0.180。
2025年：股票收益 +17.9%，宏观回归 R²=0.33；|贡献|最大为【增长】（+0.012）。分解：增长=+0.012, 通胀=-0.000, 利率=+0.004。

【十年主导因子出现次数】
主导因子
利率    4
通胀    4
增长    2

12.6 Robustness: Full-Sample β × Annual Factor Realizations

An alternative skips re-estimating β each year and uses full-sample loadings, letting only factor realizations vary by year. Comparing the two definitions reduces noise from small-sample annual regressions.

# full-sample betas
df_all = panel[factors + ["stock_ret"]].dropna
_, betas_full, _, r2_full = year_regression(df_all, factors)
print("full-sample R2=", round(r2_full, 3), "betas=", {k: round(v, 3) for k, v in betas_full.items})

rows2 = []
for y in summary.index:
    df_y = panel.loc[panel["year"] == y, factors].dropna
    c_sum = {f: float((betas_full[f] * df_y[f]).sum) for f in factors}
    winner = max(c_sum, key=lambda k: abs(c_sum[k]))
    rows2.append({"Year": y, "Dominant factor (full-sample β)": winner, **{f"Contrib_{f}": c_sum[f] for f in factors}})

robust = pd.DataFrame(rows2).set_index("Year")
cmp = summary[["Dominant factor"]].join(robust[["Dominant factor (full-sample β)"]])
cmp["Agree"] = cmp["Dominant factor"] == cmp["Dominant factor (full-sample β)"]
print(cmp.to_string)
print("Agreement rate:", f"{cmp['Agree'].mean:.0%}")
robust.to_csv(DATA_DIR / "yearly_contrib_fullbeta.csv", encoding="utf-8-sig")

full-sample R2= 0.04 betas= {'增长': np.float64(0.008), '通胀': np.float64(-0.007), '利率': np.float64(0.0)}
     主导因子 主导因子_全样本β     一致
年份                        
2016   增长        增长   True
2017   利率        利率   True
2018   通胀        通胀   True
2019   通胀        通胀   True
2020   通胀        通胀   True
2021   利率        通胀  False
2022   利率        通胀  False
2023   通胀        通胀   True
2024   利率        通胀  False
2025   增长        通胀  False
一致率: 60%

12.7 Summary

1. Classic macro factors (growth, inflation, rates, and credit when available) are enough for a coherent decade attribution story, but explanatory power varies sharply by year.
2. Who “contributes most” depends on definition: re-estimated annual β (closer to that year’s structure) vs full-sample β (more stable and comparable). Look at both.
3. Extensions: swap the left-hand side for bond returns or sector indices; or condition contributions on Chapter 10 Merrill / money–credit regimes.

Data and figures are cached by default under Data/macro_decade/.