跳到主要内容

Macro by Mark 的 Cookie.我们使用必要 Cookie 处理登录、语言和安全。经您同意,我们还会将可选 Cookie 用于保存偏好、分析,以及通过 Meta(Facebook)Pixel 衡量广告效果。请阅读我们的Cookie 政策。

Macro by MarkMacro by Mark
  • 首页
加载中…
加载中…
← 模型
Macro概念政策模型

Dynamic factor model

Extracts a small number of latent factors from many macro indicators and lets those factors drive a current-quarter read.

概述图表数学参考资料

Graph workspace

Factor decomposition with scree plot

Graph workspace

The core output of a DFM is the extracted factor or factors and their contribution to each observed variable. A dual-panel view - the factor time series with its explained variance share, and a scree plot showing eigenvalue magnitudes - captures both the dynamic behavior and the statistical justification for the chosen number of factors.

Chart axis metadata
xTimemonths · 1-120yFactor value / Eigenvaluestandardized units · -3 to 3 for factor; 0 to 30 for eigenvalue
Latent factor and its first principal
component static chart preview showing True factor, Estimated factor (first PC)
Scree plot: largest eigenvalues of the
sample covariance static chart preview showing Series 1
xTimemonths · 1-120yFactor value / Eigenvaluestandardized units · -3 to 3 for factor; 0 to 30 for eigenvalue

Current results

Variance explained (factor 1)

Explain: Variance explained (factor 1)
R12=λ1∑j=1nλjR^2_1 = \frac{\lambda_1}{\sum_{j=1}^{n} \lambda_j}R12​=∑j=1n​λj​λ1​​

Fraction of total panel variance accounted for by the first factor. Typically 20-40% for large real macro panels, which carry more than one source of common variation; this one-factor toy gives far more (about 0.86 at the defaults).

0.8566

ratio (0 to 1)

Eigenvalue ratio (1st to 2nd)

Explain: Eigenvalue ratio (1st to 2nd)
λ^1λ^2\frac{\hat{\lambda}_1}{\hat{\lambda}_2}λ^2​λ^1​​

How dominant the first factor is relative to the second. A ratio above 3-4 suggests a single dominant factor. If close to 1, two factors have similar importance.

65.56

ratio

Factor-true correlation

Explain: Factor-true correlation
∣corr(f^t,ft)∣|\text{corr}(\hat{f}_t, f_t)|∣corr(f^​t​,ft​)∣

How accurately the estimated factor tracks the true latent factor. Values above 0.9 indicate reliable extraction.

0.9962

correlation (0 to 1)

Signal-to-noise ratio

Explain: Signal-to-noise ratio
SNR=λˉ2Var(ft)σˉe2\text{SNR} = \frac{\bar{\lambda}^2 \text{Var}(f_t)}{\bar{\sigma}_e^2}SNR=σˉe2​λˉ2Var(ft​)​

Ratio of common-component variance to idiosyncratic variance for the average variable. Higher SNR means easier factor extraction.

5.63

ratio

Controls

Long swings
Explain: How long the swings last

Factor AR(1) persistence: 0.85

How long the common factor's swings last; higher values give longer business cycle swings.

High ρ\rhoρ: the factor in the top panel shows smooth, persistent cycles. Low ρ\rhoρ: the factor is choppy.

1
Explain: Strength of the factor's pull

How strongly the common factor moves the observed variables on average; higher means it explains more.

Higher λˉ\bar{\lambda}λˉ: the first eigenvalue in the scree plot towers over the rest. The factor-to-noise ratio improves.

0.8
Explain: Noise unique to each variable

How much each variable wobbles on its own; more of it hides the common factor.

Higher σe\sigma_eσe​: scree plot flattens (eigenvalues more uniform), estimated factor becomes noisier.

20
Explain: How many variables

How many observed series go into the panel; more of them give more information.

Larger n: first eigenvalue grows proportionally, factor estimate becomes more precise (closer to the true factor).

1
Explain: Size of the factor's shocks

How big the shocks hitting the common factor are; this sets the size of the swings.

Larger ση\sigma_\etaση​: wider swings in the factor time series. The factor becomes easier to distinguish from idiosyncratic noise.

Keep going

Read next

Continue with a related concept, model, or example.

Empirical model

ARDL / Bounds Testing

a contrasting take on Dynamic Factor Model

Contrast it →
Empirical model

BVAR

a contrasting take on Dynamic Factor Model

Contrast it →

继续学习

概念

  • Output and income→
  • Business cycles→
  • Labor market structure→

政策

  • Monetary policy→
  • Fiscal policy→

模型

  • Bridge model→
  • MIDAS→
  • VAR→
Macro by Mark
帮助
隐私政策使用条款伦理与合规

© 2026 Mark Jayson Nation LLC