Skip to main content

Cookies on Macro by Mark.We use essential cookies for sign-in, locale, and security. With your permission, we also use optional cookies for preferences, for analytics, and for advertising measurement with the Meta (Facebook) Pixel. Read our cookie policy.

Macro by MarkMacro by Mark
  • Home
Loading…
Loading…
← Models
MacroConceptsPolicyModels

Dynamic factor model

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

OverviewGraphMathReferences

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 →

Continue learning

Concepts

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

Policy

  • Monetary policy→
  • Fiscal policy→

Models

  • Bridge model→
  • MIDAS→
  • VAR→
Macro by Mark
Help
Privacy PolicyTerms of UseEthics and Compliance

© 2026 Mark Jayson Nation LLC