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

ETS (exponential smoothing) Model

Error / trend / seasonality state-space framework that decomposes a series into smoothed components and forecasts forward.

概述图表数学参考资料

Graph workspace

State decomposition with forecast cone

Graph workspace

The canonical ETS visualization decomposes the series into its state components (level, trend, seasonal) and shows how they combine to produce the forecast. The main panel displays the observed series with fitted values and forecast cone; smaller panels or overlays show the individual states, letting the practitioner assess whether the level is tracking the data, the trend is reasonable, and the seasonal shape is stable.

Chart axis metadata
xTime (period index)integer period · t=1t = 1t=1 to T+HT + HT+H, where T≈80T \approx 80T≈80 and H≈24H \approx 24H≈24ySeries levelindex units · Mean about 100, range 60 to 140 depending on trend and seasonal amplitude

Shaded ranges come from assumed parameters in a simulated example. They are not estimated from observed data.

Time-series forecast with illustrative ranges: Series level (index units) by Time (period index) static chart preview showing History, Forecast
xTime (period index)integer period · t=1t = 1t=1 to T+HT + HT+H, where T≈80T \approx 80T≈80 and H≈24H \approx 24H≈24ySeries levelindex units · Mean about 100, range 60 to 140 depending on trend and seasonal amplitude

Current results

1-step forecast

Explain: 1-step forecast
y^T+1=lT+ϕbT+sT+1−s\hat{y}_{T+1} = l_T + \phi b_T + s_{T+1-s}y^​T+1​=lT​+ϕbT​+sT+1−s​

Next period's point forecast: current level plus damped trend plus the relevant seasonal factor.

100.09

index level

Current level

Explain: Current level
lTl_TlT​

The model's estimate of the underlying de-trended, de-seasonalized series value right now.

104.27

index level

Damped trend at h=12

Explain: Damped trend at h=12
ϕ12bT=(ϕ+ϕ2+⋯+ϕ12)bT\phi_{12} b_T = (\phi + \phi^2 + \cdots + \phi^{12}) b_Tϕ12​bT​=(ϕ+ϕ2+⋯+ϕ12)bT​

Cumulative trend contribution after 12 periods. For ϕ=0.9\phi=0.9ϕ=0.9, the damped contribution is about 54% of the undamped linear trend over 12 periods.

-1.14

index units, summed over the 12 periods

95% interval width at h=12

Explain: 95% interval width at h=12
2×1.96×σk122 \times 1.96 \times \sigma \sqrt{k_{12}}2×1.96×σk12​​

Forecast uncertainty one year ahead. The khk_hkh​ multiplier depends on the smoothing parameters and grows with horizon.

15.11

index units

Controls

Balanced
Explain: How fast the level updates

Level smoothing: 0.3

How much the model trusts the newest observation versus the historical level.

High α\alphaα: fitted values hug the data closely, level state is jagged. Low α\alphaα: fitted values are smooth, level state is a slow-moving average.

Trend drifts slowly
Explain: How fast the trend updates

Trend smoothing: 0.05

How quickly the trend reacts to new data; usually kept small for stability.

High β\betaβ: trend state fluctuates visibly period-to-period. Low β\betaβ: trend state is nearly constant, producing a smooth forecast trajectory.

Slowly evolving season
Explain: Seasonal pattern flexibility

Seasonal smoothing: 0.1

How quickly the repeating seasonal shape is allowed to change over time.

High γ\gammaγ: seasonal sub-panel shows a pattern that shifts shape over time. Low γ\gammaγ: seasonal sub-panel shows a fixed repeating wave.

Growth fades gradually
Explain: How fast growth fades

Trend damping: 0.9

How much strength the trend loses each period; at one it never fades.

ϕ<1\phi < 1ϕ<1: the forecast path curves toward a horizontal asymptote. ϕ=1\phi = 1ϕ=1: the forecast path extrapolates linearly. The difference is most visible at long horizons.

Moderately noisy
Explain: Size of prediction errors

Innovation std dev: 2

How big one-step errors are, setting series noisiness and forecast interval width.

Larger σ\sigmaσ makes the historical series noisier and widens all forecast intervals proportionally.

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
帮助
隐私政策使用条款伦理与合规

© 2026 Mark Jayson Nation LLC