Saltar al contenido principal

Cookies en Macro by Mark.Usamos cookies esenciales para inicio de sesión, idioma y seguridad. Con tu permiso, también usamos cookies opcionales de preferencias, para analítica y para medición publicitaria con el píxel de Meta (Facebook). Lee nuestra política de cookies.

Macro by MarkMacro by Mark
  • Inicio
Cargando…
Cargando…
← Modelos
MacroConceptosPolíticaModelos

MIDAS Model

Mixed-data sampling regressions that handle indicators arriving at different frequencies inside a single forecasting equation.

ResumenGráficoMatemáticasReferencias

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.

The Midas Regression Model

Let ytLy_t^LytL​ be observed at frequency LLL (quarterly) and xτHx_{\tau}^HxτH​ at frequency HHH (monthly). The frequency ratio m=3m = 3m=3. Within each low-frequency period ttt, there are mmm high-frequency observations. Indexing: the most recent monthly observation in quarter ttt is xtmHx_{tm}^HxtmH​, the previous is xtm−1Hx_{tm-1}^Hxtm−1H​, and so on back to xtm−KHx_{tm-K}^Hxtm−KH​ for KKK total lags.

The MIDAS regression replaces the temporally aggregated predictor of the bridge equation with a weighted sum of high-frequency lags: ytL=α+βB(L1/m;θ)xtmH+εty_t^L = \alpha + \beta B(L^{1/m}; \theta) x_{tm}^H + \varepsilon_tytL​=α+βB(L1/m;θ)xtmH​+εt​, where B(L1/m;θ)B(L^{1/m}; \theta)B(L1/m;θ) is a lag polynomial in the high-frequency lag operator, with coefficients parameterized by θ\thetaθ.

ytL=α+β∑k=0Kw(k;θ)⋅xtm−kH+εt,εt∼(0,σε2)y_t^L = \alpha + \beta \sum_{k=0}^{K} w(k; \boldsymbol{\theta}) \cdot x_{tm-k}^H + \varepsilon_t, \quad \varepsilon_t \sim (0, \sigma_\varepsilon^2)ytL​=α+βk=0∑K​w(k;θ)⋅xtm−kH​+εt​,εt​∼(0,σε2​)
ytLy_t^LytL​α\alphaαβ\betaβKKKw(k;θ)w(k; \boldsymbol{\theta})w(k;θ)

The basic MIDAS regression. The weighted sum replaces the temporal aggregation of bridge equations.

w(k;θ1,θ2)=exp⁡(θ1k+θ2k2)∑j=0Kexp⁡(θ1j+θ2j2)w(k; \theta_1, \theta_2) = \frac{\exp(\theta_1 k + \theta_2 k^2)}{\sum_{j=0}^{K} \exp(\theta_1 j + \theta_2 j^2)}w(k;θ1​,θ2​)=∑j=0K​exp(θ1​j+θ2​j2)exp(θ1​k+θ2​k2)​
θ1\theta_1θ1​θ2\theta_2θ2​KKK

Exponential Almon weighting function. Two parameters control shape: θ1\theta_1θ1​ for tilt, θ2\theta_2θ2​ for curvature.

w(k;a,b)=xka−1(1−xk)b−1∑j=0Kxja−1(1−xj)b−1,xk=k+1K+2w(k; a, b) = \frac{x_k^{a-1}(1 - x_k)^{b-1}}{\sum_{j=0}^{K} x_j^{a-1}(1 - x_j)^{b-1}}, \quad x_k = \frac{k+1}{K+2}w(k;a,b)=∑j=0K​xja−1​(1−xj​)b−1xka−1​(1−xk​)b−1​,xk​=K+2k+1​
KKK

Beta polynomial weighting function, evaluated on the interior grid xk=(k+1)/(K+2)x_k = (k+1)/(K+2)xk​=(k+1)/(K+2) so the density stays finite at the support endpoints when a<1a < 1a<1 or b<1b < 1b<1. Can produce U-shapes, declining, or hump-shaped profiles depending on (a,b)(a, b)(a,b).

ytL=α+∑k=0Kβkxtm−kH+εty_t^L = \alpha + \sum_{k=0}^{K} \beta_k x_{tm-k}^H + \varepsilon_tytL​=α+k=0∑K​βk​xtm−kH​+εt​
ytLy_t^LytL​α\alphaαKKK

Unrestricted MIDAS (U-MIDAS). No weighting function - OLS estimates all K+1 coefficients.

ytL=α+ρyt−1L+β∑k=0Kw(k;θ)⋅xtm−kH+εty_t^L = \alpha + \rho y_{t-1}^L + \beta \sum_{k=0}^{K} w(k; \boldsymbol{\theta}) \cdot x_{tm-k}^H + \varepsilon_tytL​=α+ρyt−1L​+βk=0∑K​w(k;θ)⋅xtm−kH​+εt​
ytLy_t^LytL​α\alphaαβ\betaβKKKw(k;θ)w(k; \boldsymbol{\theta})w(k;θ)

MIDAS-AR: adds an autoregressive term for the low-frequency target. Handles persistent dynamics in the target.

Align the high-frequency and low-frequency time indices

Correct temporal alignment is critical - an off-by-one error shifts the entire weight profile
Quarter t↔months (3t−2,3t−1,3t).xtmH=x3tH,  xtm−1H=x3t−1H,…\text{Quarter } t \leftrightarrow \text{months } (3t-2, 3t-1, 3t). \quad x_{tm}^H = x_{3t}^H, \; x_{tm-1}^H = x_{3t-1}^H, \ldotsQuarter t↔months (3t−2,3t−1,3t).xtmH​=x3tH​,xtm−1H​=x3t−1H​,…

Choose the number of high-frequency lags K

K determines how far back the monthly information extends
K=m⋅q−1 for q quarters of monthly lags. K=11 for 4 quarters.K = m \cdot q - 1 \text{ for } q \text{ quarters of monthly lags. } K = 11 \text{ for 4 quarters.}K=m⋅q−1 for q quarters of monthly lags. K=11 for 4 quarters.
KKKmmm

Select the weighting function family

Different families offer different flexibility-parsimony tradeoffs
Exponential Almon (2 params), Beta polynomial (2 params), step function (1 param per step), or unrestricted (K+1 params)\text{Exponential Almon (2 params), Beta polynomial (2 params), step function (1 param per step), or unrestricted (K+1 params)}Exponential Almon (2 params), Beta polynomial (2 params), step function (1 param per step), or unrestricted (K+1 params)

Construct the weighted predictor ztz_tzt​

The MIDAS regression reduces to ytL=α+βzt+εty_t^L = \alpha + \beta z_t + \varepsilon_tytL​=α+βzt​+εt​ once ztz_tzt​ is computed
zt(θ)=∑k=0Kw(k;θ)⋅xtm−kHz_t(\boldsymbol{\theta}) = \sum_{k=0}^{K} w(k; \boldsymbol{\theta}) \cdot x_{tm-k}^Hzt​(θ)=k=0∑K​w(k;θ)⋅xtm−kH​
θ\boldsymbol{\theta}θKKKw(k;θ)w(k; \boldsymbol{\theta})w(k;θ)

Verify that flat weights recover the bridge equation

MIDAS nests bridge equations as a special case
θ1=θ2=0,  K=m−1⇒w(k)=1/m⇒zt=1m∑k=0m−1xtm−kH=xˉQ\theta_1 = \theta_2 = 0, \; K = m-1 \Rightarrow w(k) = 1/m \Rightarrow z_t = \frac{1}{m}\sum_{k=0}^{m-1} x_{tm-k}^H = \bar{x}_Qθ1​=θ2​=0,K=m−1⇒w(k)=1/m⇒zt​=m1​k=0∑m−1​xtm−kH​=xˉQ​
θ1\theta_1θ1​θ2\theta_2θ2​KKKmmm

Note the identification restriction

The product βw(k)\beta w(k)βw(k) is identified, but β\betaβ and www separately require a normalization
Normalization: ∑k=0Kw(k;θ)=1⇒β measures the total effect of a unit increase in all monthly lags\text{Normalization: } \sum_{k=0}^{K} w(k; \boldsymbol{\theta}) = 1 \Rightarrow \beta \text{ measures the total effect of a unit increase in all monthly lags}Normalization: k=0∑K​w(k;θ)=1⇒β measures the total effect of a unit increase in all monthly lags
KKKw(k;θ)w(k; \boldsymbol{\theta})w(k;θ)β\betaβ

Why MIDAS instead of more OLS lags?

With m=3m = 3m=3 and K=11K = 11K=11, U-MIDAS has 12 free coefficients plus the intercept. A quarterly sample of T=80T = 80T=80 can handle this. But with daily-to-monthly frequency (m=22),K=65(m = 22), K = 65(m=22),K=65 for 3 months of daily history means 66 free coefficients - OLS breaks down. The parametric weighting function reduces this to 2-3 parameters regardless of KKK.

Notations

L1/mL^{1/m}L1/m

High-frequency lag operator

B(L1/m;θ)B(L^{1/m}; \boldsymbol{\theta})B(L1/m;θ)

MIDAS lag polynomial

About these symbols

High-frequency lag operator: High-frequency lag operator: L1/mxtmH=xtm−1HL^{1/m} x_{tm}^H = x_{tm-1}^HL1/mxtmH​=xtm−1H​

MIDAS lag polynomial: MIDAS lag polynomial in the high-frequency lag operator

    Transition checks

    Forecast origin
    y^T+h∣T=ET[yT+h],h=1,2,…\widehat{y}_{T+h|T}=\mathbb{E}_T[y_{T+h}],\qquad h=1,2,\ldotsy​T+h∣T​=ET​[yT+h​],h=1,2,…

    Conditioning on information at T separates the sample from the projection. The forecast can be smooth even when later realized data will not be.

    The bridge point is shared so the eye can see the handoff. After that point, the red line is a model forecast, not observed history.

    Do not score the forecast segment as realized data or read it as a policy promise.

    Uncertainty propagation
    Var⁡T(yT+h)=σ2∑j=0h−1ψj2\operatorname{Var}_T(y_{T+h})=\sigma^2\sum_{j=0}^{h-1}\psi_j^2VarT​(yT+h​)=σ2j=0∑h−1​ψj2​

    Each future innovation contributes through its impulse weight ψj\psi_jψj​. Unless the future is degenerate, longer horizons carry more conditional variance.

    The band widens when shocks accumulate through the model recursion.

    Do not treat the band edge as an alternative deterministic scenario.

    Variable timing

    Observed history

    yt∈ITy_t\in\mathcal{I}_Tyt​∈IT​

    The history segment is realized data available at the forecast origin.

    Conditional mean

    y^T+h∣T=E[yT+h∣IT]\widehat{y}_{T+h|T}=\mathbb{E}[y_{T+h}\mid\mathcal{I}_T]y​T+h∣T​=E[yT+h​∣IT​]

    The forecast path is a conditional expectation, so it can be smooth while future realizations remain volatile.

    Prediction band

    σh2=Var⁡(yT+h∣IT)\sigma_h^2=\operatorname{Var}(y_{T+h}\mid\mathcal{I}_T)σh2​=Var(yT+h​∣IT​)

    Band width is horizon-specific uncertainty, not a second deterministic transition.

    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
    Ayuda
    Política de privacidadTérminos de usoÉtica y cumplimiento

    © 2026 Mark Jayson Nation LLC