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

ARIMA Model

Autoregressive integrated moving average - the workhorse univariate forecaster for stationary or differenced macro series.

概述图表数学参考资料

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.

Model Specification and Notation

The ARIMA(p,d,q)\text{ARIMA}(p, d, q)ARIMA(p,d,q) model operates on a scalar time series yty_tyt​ observed at equally spaced dates t=1,2,…,Tt = 1, 2, \ldots, Tt=1,2,…,T. The model assumes that after applying the difference operator ddd times, the resulting series is a covariance-stationary ARMA(p,q)\text{ARMA}(p, q)ARMA(p,q) process. All notation that follows treats the differenced series as the fundamental object of estimation.

Define the lag operator LLL by Lyt=yt−1L y_t = y_{t-1}Lyt​=yt−1​ and the difference operator by Δ=1−L\Delta = 1 - LΔ=1−L. The d-differenced series is wt=Δdyt=(1−L)dytw_t = \Delta^d y_t = (1 - L)^d y_twt​=Δdyt​=(1−L)dyt​. For most macroeconomic applications, ddd is either 0 (the raw series is stationary) or 1 (the series has a unit root in levels but is stationary in first differences). Second differencing is rare and signals that the series has two unit roots - typically an artifact of over-differencing or a genuinely explosive process that other models handle better.

ϕ(L) (1−L)d yt=c+θ(L) εt\phi(L)\,(1-L)^{d}\, y_t = c + \theta(L)\,\varepsilon_tϕ(L)(1−L)dyt​=c+θ(L)εt​
ϕ(L)\phi(L)ϕ(L)(1−L)d(1 - L)^d(1−L)dyty_tyt​cccθ(L)\theta(L)θ(L)εt\varepsilon_tεt​

The compact ARIMA(p,d,q)\text{ARIMA}(p, d, q)ARIMA(p,d,q) representation: the AR polynomial acts on the d-differenced series, and the MA polynomial drives the innovations.

ϕ(L)=1−ϕ1L−ϕ2L2−⋯−ϕpLp\phi(L) = 1 - \phi_1 L - \phi_2 L^2 - \cdots - \phi_p L^pϕ(L)=1−ϕ1​L−ϕ2​L2−⋯−ϕp​Lp
ϕ(L)\phi(L)ϕ(L)ϕ1\phi_1ϕ1​LLL

The AR polynomial of order p. Stationarity of the ARMA representation requires all p roots of this polynomial to lie strictly outside the unit circle in the complex plane.

θ(L)=1+θ1L+θ2L2+⋯+θqLq\theta(L) = 1 + \theta_1 L + \theta_2 L^2 + \cdots + \theta_q L^qθ(L)=1+θ1​L+θ2​L2+⋯+θq​Lq
θ(L)\theta(L)θ(L)θ1\theta_1θ1​LLL

The MA polynomial of order q. Invertibility requires all q roots to lie strictly outside the unit circle, ensuring the AR(infinity) representation converges.

wt=c+ϕ1wt−1+⋯+ϕpwt−p+εt+θ1εt−1+⋯+θqεt−qw_t = c + \phi_1 w_{t-1} + \cdots + \phi_p w_{t-p} + \varepsilon_t + \theta_1 \varepsilon_{t-1} + \cdots + \theta_q \varepsilon_{t-q}wt​=c+ϕ1​wt−1​+⋯+ϕp​wt−p​+εt​+θ1​εt−1​+⋯+θq​εt−q​
wtw_twt​cccϕ1\phi_1ϕ1​εt\varepsilon_tεt​θ1\theta_1θ1​

The expanded ARMA(p,q)\text{ARMA}(p, q)ARMA(p,q) form on the differenced series, with constant ccc representing drift.

Notations

yty_tyt​

Observed series

wtw_twt​

Differenced series

LLL

Lag operator

ϕ(L)\phi(L)ϕ(L)

AR polynomial

θ(L)\theta(L)θ(L)

MA polynomial

εt\varepsilon_tεt​

Innovation

About these symbols

Observed series:

Differenced series: wt=(1−L)dytw_t = (1-L)^d y_twt​=(1−L)dyt​

Lag operator: Lkyt=yt−kL^k y_t = y_{t-k}Lkyt​=yt−k​

AR polynomial:

MA polynomial:

Innovation: WN(0,σ2)\text{WN}(0, \sigma^2)WN(0,σ2)

    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​
    σ2\sigma^2σ2

    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​
    yty_tyt​

    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.

    Empirical model

    ARDL / Bounds Testing

    a contrasting take on ARIMA

    Contrast it →
    Empirical model

    Bridge Equations

    a contrasting take on ARIMA

    Contrast it →

    继续学习

    概念

    • Real and nominal→
    • Business cycles→
    • Prices and inflation→

    政策

    • Monetary policy→
    • Fiscal policy→

    模型

    • SARIMA→
    • ETS→
    • VAR→
    Macro by Mark
    帮助
    隐私政策使用条款伦理与合规

    © 2026 Mark Jayson Nation LLC