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

Satellite models

Smaller models that hang off the core macroeconometric system to handle housing, energy, financial conditions, or other special topics.

OverviewGraphMathReferences

Understand the mathematics

Follow the same equations and assumptions one section at a time or in a continuous reading.

Hover or focus a symbol defined in this model to find its exact matches. Select it to keep the highlight; Escape clears the selection.

Satellite Architecture and Conditional Forecasting

The satellite architecture decomposes the full economic system into a core and a set of peripheral models. The core handles aggregate dynamics; each satellite handles one domain. The decomposition is valid under the assumption that the satellite variables do not Granger-cause the core variables.

Formally, partition the full state vector into core variables ztz_tzt​ and satellite variables yt(s)y_t^{(s)}yt(s)​. If y(s)y^{(s)}y(s) does not Granger-cause zzz, then the core can be estimated and forecast independently, and the satellite can be estimated conditioning on zzz. This is a block-recursive structure.

The conditional forecast from a satellite model is the expected value of y(s)y^{(s)}y(s) given the core's projected path for zzz. The satellite's forecast error decomposes into two parts: the error from the satellite model itself and the error inherited from the core's forecast of zzz.

Full system: [ztyt(s)]=[A11A12A21A22][zt−1yt−1(s)]+[utzuts]\text{Full system: } \begin{bmatrix} z_t \\ y_t^{(s)} \end{bmatrix} = \begin{bmatrix} A_{11} & A_{12} \\ A_{21} & A_{22} \end{bmatrix} \begin{bmatrix} z_{t-1} \\ y_{t-1}^{(s)} \end{bmatrix} + \begin{bmatrix} u_t^z \\ u_t^s \end{bmatrix}Full system: [zt​yt(s)​​]=[A11​A21​​A12​A22​​][zt−1​yt−1(s)​​]+[utz​uts​​]
yt(s)y_t^{(s)}yt(s)​A12A_{12}A12​A21A_{21}A21​

Full VAR(1) representation of core and satellite variables. The satellite architecture sets A12=0A_{12} = 0A12​=0: satellite variables do not enter the core's law of motion.

Block-recursive: zt=A11zt−1+utz,yt(s)=A21zt−1+A22yt−1(s)+uts\text{Block-recursive: } z_t = A_{11} z_{t-1} + u_t^z, \quad y_t^{(s)} = A_{21} z_{t-1} + A_{22} y_{t-1}^{(s)} + u_t^sBlock-recursive: zt​=A11​zt−1​+utz​,yt(s)​=A21​zt−1​+A22​yt−1(s)​+uts​
yt(s)y_t^{(s)}yt(s)​A21A_{21}A21​

With A12=0A_{12} = 0A12​=0, the core evolves independently. The satellite conditions on lagged core variables (and its own lags) plus domain shocks.

Step 1

Granger causality tests determine whether the one-way architecture is justified. If the satellite variable Granger-causes the core (A12≠0)(A_{12} \neq 0)(A12​=0), the one-way linkage misses feedback. Standard F-tests or likelihood ratio tests applied to the unrestricted VAR.
Test: H0:A12=0 (Granger non-causality from satellite to core)\text{Test: } H_0: A_{12} = 0 \text{ (Granger non-causality from satellite to core)}Test: H0​:A12​=0 (Granger non-causality from satellite to core)
A12A_{12}A12​

Step 2

Because A12=0A_{12} = 0A12​=0, the core equation is correctly specified without the satellite variables. OLS on the core block is consistent. This is the key computational advantage: the core can be estimated, solved, and updated without touching the satellites.
Estimate core: A^11 from the core block alone (OLS consistent)\text{Estimate core: } \hat{A}_{11} \text{ from the core block alone (OLS consistent)}Estimate core: A^11​ from the core block alone (OLS consistent)

Step 3

The satellite equation is estimated using observed core variables as regressors. Because zzz is exogenous to the satellite (A12=0)(A_{12} = 0)(A12​=0), OLS is consistent for A21A_{21}A21​ and A22A_{22}A22​.
Estimate satellite: A^21,A^22 from yt(s) on (zt−1,yt−1(s))\text{Estimate satellite: } \hat{A}_{21}, \hat{A}_{22} \text{ from } y_t^{(s)} \text{ on } (z_{t-1}, y_{t-1}^{(s)})Estimate satellite: A^21​,A^22​ from yt(s)​ on (zt−1​,yt−1(s)​)
yt(s)y_t^{(s)}yt(s)​

Step 4

Run the core model forward to produce the conditioning path z^\hat{z}z^ for horizons 1 through HHH. This path is treated as known by the satellite models.
Core forecast: z^T+h∣T=A^11hzT\text{Core forecast: } \hat{z}_{T+h|T} = \hat{A}_{11}^h z_TCore forecast: z^T+h∣T​=A^11h​zT​
z^T+h∣T\hat{z}_{T+h|T}z^T+h∣T​

Step 5

The satellite iterates forward using the core's projected path. At each horizon, the satellite uses the core's forecast (not the actual future realization) as its conditioning variable. This is the conditional forecast.
Satellite conditional forecast: y^T+h∣T(s)=A^21z^T+h−1∣T+A^22y^T+h−1∣T(s)\text{Satellite conditional forecast: } \hat{y}_{T+h|T}^{(s)} = \hat{A}_{21} \hat{z}_{T+h-1|T} + \hat{A}_{22} \hat{y}_{T+h-1|T}^{(s)}Satellite conditional forecast: y^​T+h∣T(s)​=A^21​z^T+h−1∣T​+A^22​y^​T+h−1∣T(s)​

Step 6

The satellite's forecast error has three sources: (1) estimation error in satellite parameters, (2) the core model's forecast error propagated through A21A_{21}A21​, and (3) the satellite's own shock. Source (2) is typically the largest contributor at longer horizons. This means satellite forecast quality is bounded by core forecast quality.
Forecast error: eT+h(s)=(A21−A^21)zT+h−1+A21(zT+h−1−z^T+h−1∣T)+uT+hs+…\text{Forecast error: } e_{T+h}^{(s)} = (A_{21} - \hat{A}_{21}) z_{T+h-1} + A_{21}(z_{T+h-1} - \hat{z}_{T+h-1|T}) + u_{T+h}^s + \ldotsForecast error: eT+h(s)​=(A21​−A^21​)zT+h−1​+A21​(zT+h−1​−z^T+h−1∣T​)+uT+hs​+…
A21A_{21}A21​

When feedback matters

The housing-to-macro feedback channel was dramatically demonstrated in 2007-2009. House price declines destroyed household wealth, reducing consumption (wealth effect), triggered mortgage defaults that impaired bank balance sheets, tightening credit supply (financial accelerator), and reduced construction employment (direct GDP component). A housing satellite with one-way linkage from macro to housing would have missed all of these feedback channels. The lesson: for systemically important sectors, the one-way satellite architecture underestimates tail risk.

Notations

A12A_{12}A12​

Satellite-to-core coefficients

A21A_{21}A21​

Core-to-satellite coefficients

z^T+h∣T\hat{z}_{T+h|T}z^T+h∣T​

Conditioning path

About these symbols

Satellite-to-core coefficients: Coefficient matrix on satellite variables in the core equation. Zero under the satellite architecture.

Core-to-satellite coefficients: Coefficient matrix on core variables in the satellite equation. This is the bridge/transmission parameter.

Conditioning path: Core model's h-step-ahead forecast of macro variables, used as conditioning path for the satellite.

    Transition checks

    Sampled dynamics
    zt=G(zt−1;θ)+ηt,yt=Hzt+εtz_t=G(z_{t-1};\theta)+\eta_t,\qquad y_t=H z_t+\varepsilon_tzt​=G(zt−1​;θ)+ηt​,yt​=Hzt​+εt​

    The state is updated at observed dates. A line between dates is a display convention unless the model specifies behavior inside the interval.

    Read adjacent points as adjacent sample calculations. Smoothness is a property of the estimator, not of the chart line.

    Do not infer within-period behavior from a sampled estimator.

    Variable timing

    Latent state

    zt=G(zt−1;θ)+ηtz_t=G(z_{t-1};\theta)+\eta_tzt​=G(zt−1​;θ)+ηt​

    The state updates at sample dates. The visible line is only as smooth as the estimated state sequence.

    Observation equation

    yt=Hzt+εty_t=Hz_t+\varepsilon_tyt​=Hzt​+εt​

    Observed series can move differently from latent states because measurement noise and loadings matter.

    Keep going

    Read next

    Continue with a related concept, model, or example.

    Theory model

    IS-LM Model

    a well-connected theoretical model two steps from here

    New angle: theoretical model →
    Macro by Mark
    Help
    Privacy PolicyTerms of UseEthics and Compliance

    © 2026 Mark Jayson Nation LLC