传统宏观因子、宏观周期的高维度体系构建 2025 年 12 月 08 日 金 融 工 程 高维宏观周期驱动风格、行业月报(2025/11):经济 景气下行、通胀细分项下行看好大盘价值风格 投资要点： 金 融 工 程 定 期 报 告 宏观因子变量的构建：将宏观指数分别对宽基指数、代理宏观变 量做回归，选取 t 值显著的细分宏观变量，用过去一年标准差倒数加权 构建宏观因子变量。采用单边 HP 滤波器对宏观经济数据进行调整，消 除短期波动对长期趋势判断的影响。基于滤波变量，分别用因子动量 划分宏观趋势（上行、下行）和用时序百分位划分宏观状态（高、中、 低位）。 宏观因子升维的必要性：宏观因子 A 对宽基、风格和行业的价格 传导在 A 的不同边际变化不一致，且宏观因子 A 在宏观因子 B 的不同 状态下驱动宽基、风格和行业的收益方向也不同。同一状态及其边际 变化所对应的周期混乱，我们需要将宏观变量的边际与状态结合，综 合考虑宏观变量的变化趋势和所处的时序排位。 多信号驱动下的指数择时、风格轮动 小盘全指择时：在库存处于中等向上水平时预测值最高，因此推 荐配置中证全指。 2012 年 1 月末起至 2025 年 11 ...

- The artificial intelligence global asset allocation model applies machine learning to asset allocation problems, using factor investment ideas to score and rank assets, ultimately constructing a monthly quantitative equal-weighted strategy for global asset allocation[38][39][40] - The dynamic macroeconomic event factor-based stock-bond rotation strategy includes three risk preference models (conservative, balanced, and aggressive), utilizing macro timing modules and risk budgeting frameworks to determine stock and bond weights[43][44][45] - The dividend style timing model uses 10 indicators from economic growth and monetary liquidity dimensions, constructing a timing strategy for the dividend index, which shows significant stability improvement compared to the CSI Dividend Total Return Index[51][54][55] Model Backtesting Results - Artificial intelligence global asset allocation model: annualized return 38.76%, Sharpe ratio 1.07, maximum drawdown -6.56%, year-to-date return 6.81%[39][40][42] - Dynamic macroeconomic event factor-based stock-bond rotation strategy: aggressive model annualized return 20.14%, Sharpe ratio 1.30, maximum drawdown -13.72%, year-to-date return 14.42%; balanced model annualized return 10.92%, Sharpe ratio 1.19, maximum drawdown -6.77%, year-to-date return 4.13%; conservative model annualized return 5.94%, Sharpe ratio 1.50, maximum drawdown -3.55%, year-to-date return 0.97%[43][49][50] - Dividend style timing model: annualized return 16.52%, Sharpe ratio 1.07, maximum drawdown -13.77%, year-to-date return 0%[51][54][55]

- The report includes a global asset allocation model based on artificial intelligence, which uses machine learning to score and rank various assets for monthly equal-weighted allocation strategy[30][31] - The global asset allocation model suggests weights for May: government bond index (66.09%), Nasdaq index (17.59%), German DAX index (13.83%), and Nikkei 225 (2.49%)[30] - Historical performance of the global asset allocation model from January 2021 to April 2025 shows an annualized return of 13.76%, Sharpe ratio of 0.75, maximum drawdown of 16.53%, and excess annualized return of 9.02%[30][36] - The dynamic macro event factor-based stock-bond rotation strategy includes three different risk preference models: conservative, balanced, and aggressive[37] - The stock-bond allocation models for April show stock weights of 45% for aggressive, 13.82% for balanced, and 0% for conservative[37][39] - Historical performance of the stock-bond allocation models from January 2005 to April 2025 shows annualized returns of 19.93% for aggressive, 11.00% for balanced, and 6.06% for conservative[37][44] - The dividend timing model uses economic growth and monetary liquidity indicators to construct a timing strategy for the dividend index, showing an annualized return of 15.84%, maximum drawdown of -21.70%, and Sharpe ratio of 0.89[45][49] - The dividend timing model's recommended position for April is 0%, with most economic growth indicators showing bearish signals and cautious monetary liquidity signals[45] Model Performance Metrics - Global asset allocation model: annualized return 13.76%, Sharpe ratio 0.75, maximum drawdown 16.53%[30][36] - Stock-bond allocation models: annualized returns 19.93% (aggressive), 11.00% (balanced), 6.06% (conservative)[37][44] - Dividend timing model: annualized return 15.84%, Sharpe ratio 0.89, maximum drawdown -21.70%[45][49]

- The global asset allocation model uses machine learning to score and rank assets based on factor investment principles, constructing a monthly quantitative equal-weight strategy for global asset allocation[39][43][44] - The model's historical performance from January 2021 to March 2025 shows an annualized return of 6.45%, Sharpe ratio of 1.01, maximum drawdown of 6.66%, and excess annualized return of 1.28%, outperforming the benchmark across all dimensions[39][44][45] - The dynamic macro event factor-based stock-bond rotation strategy includes three risk preference models (conservative, balanced, aggressive), with April stock weights of 0%, 13.73%, and 25%, respectively[45][46][47] - The macro timing module and risk budget framework signal strengths for April are 50% for monetary liquidity and 0% for economic growth[45][46][48] - Historical performance of the stock-bond rotation strategy from January 2005 to March 2025 shows annualized returns of 20.02% (aggressive), 11.02% (balanced), and 6.03% (conservative), all outperforming the benchmark[45][51][47] - The dividend timing model recommends a 100% allocation to the CSI Dividend Index for April, with economic growth indicators mostly bearish and monetary liquidity signals positive[53][54][52] - The dividend timing strategy achieves an annualized return of 16.86%, maximum drawdown of -21.22%, and Sharpe ratio of 0.95, significantly improving stability compared to the CSI Dividend Total Return Index[53][54][52]