Aller au contenu principal

Cookies sur Macro by Mark.Nous utilisons des cookies essentiels pour la connexion, la langue et la sécurité. Avec votre accord, nous ajoutons des cookies optionnels de préférences, d’analyse et de mesure publicitaire avec le pixel Meta (Facebook). Lisez notre politique relative aux cookies.

Macro by MarkMacro by Mark
  • Accueil
Chargement…
Chargement…
← Modèles
MacroConceptsPolitiqueModèles

SARIMA Model

Seasonal ARIMA - extends ARIMA with explicit seasonal AR, differencing, and MA terms for series with calendar structure.

Vue d’ensembleGraphiqueMathématiquesRéférences

Graph workspace

Seasonal time-series forecast cone with seasonal overlay

Graph workspace

The canonical visualization for SARIMA adds a seasonal decomposition to the standard forecast cone. The primary panel shows the full series with historical data, fitted values, and the forecast cone. A secondary overlay or panel shows the seasonal component extracted from the model, making the repeating pattern visible. This lets the practitioner see both the overall forecast trajectory and whether the seasonal shape looks reasonable.

Chart axis metadata
xTime (period index)integer period · t=1t = 1t=1 to T+2sT + 2sT+2s, where T≈96T \approx 96T≈96 (8 years of monthly data) and the forecast horizon is H=2s=24H = 2s = 24H=2s=24ySeries levelindex units · Starts at 100 with a seasonal amplitude of about 1 to 20 units, scaling with σ\sigmaσ and growing with Φ1\Phi_1Φ1​ and as Θ1\Theta_1Θ1​ nears 0 (about 2.8 at the defaults). With d=D=1d = D = 1d=D=1 the level is a random walk, so it wanders from 100 rather than reverting to it, and a non-zero drift ccc bends it away quadratically.

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+2sT + 2sT+2s, where T≈96T \approx 96T≈96 (8 years of monthly data) and the forecast horizon is H=2s=24H = 2s = 24H=2s=24ySeries levelindex units · Starts at 100 with a seasonal amplitude of about 1 to 20 units, scaling with σ\sigmaσ and growing with Φ1\Phi_1Φ1​ and as Θ1\Theta_1Θ1​ nears 0 (about 2.8 at the defaults). With d=D=1d = D = 1d=D=1 the level is a random walk, so it wanders from 100 rather than reverting to it, and a non-zero drift ccc bends it away quadratically.

Current results

1-step forecast

Explain: 1-step forecast
y^T+1∣T\hat{y}_{T+1|T}y^​T+1∣T​

Next-period point forecast, using the full SARIMA recursion with all in-sample residual MA corrections.

95.10

index level

Seasonal-ahead forecast

Explain: Seasonal-ahead forecast
y^T+s∣T\hat{y}_{T+s|T}y^​T+s∣T​

The forecast for the same calendar month one year ahead. Tests whether the model's seasonal pattern is reasonable.

93.27

index level

95% interval width at h = s

Explain: 95% interval width at h = s
2×1.96×σ∑j=0s−1Ψj22 \times 1.96 \times \sigma \sqrt{\sum_{j=0}^{s-1}\Psi_j^2}2×1.96×σ∑j=0s−1​Ψj2​​

Uncertainty one full seasonal cycle ahead, after undoing seasonal and ordinary differencing in the impulse weights.

11.88

index units

Seasonal amplitude

Explain: Seasonal amplitude
max⁡ky^T+k−min⁡ky^T+k,k=1,…,s\max_k \hat{y}_{T+k} - \min_k \hat{y}_{T+k}, \quad k = 1, \ldots, smaxk​y^​T+k​−mink​y^​T+k​,k=1,…,s

Peak-to-trough range in the first forecasted seasonal cycle. Measures how pronounced the model thinks the seasonal swing is.

2.84

index units

Controls

Some carryover
Explain: Carryover from last month

Non-seasonal AR(1): 0.3

Sets how much of last month's gap from the usual seasonal pattern carries into this month.

Higher ϕ1\phi_1ϕ1​ makes the series smoother within each seasonal cycle. The forecast path adjusts more gradually after the forecast origin.

Damps recent shocks
Explain: Weight on last period's error

Non-seasonal MA(1): -0.4

How much last period's forecast error counts; negative values soften the reaction to recent shocks.

Changing θ1\theta_1θ1​ shifts the first forecast step relative to the last observation and affects how quickly the within-season forecast converges to the seasonal pattern.

No year-to-year link
Explain: Carryover from last year

Seasonal AR(1): 0

Sets how much this January depends on last January, beyond the repeating yearly pattern.

Positive Φ1\Phi_1Φ1​ makes year-over-year deviations persist: a hot December predicts a hot next December. The seasonal shape in the forecast slowly adjusts rather than snapping to the historical average.

Absorbs seasonal shocks
Explain: Weight on last year's error

Seasonal MA(1): -0.6

How much last year's seasonal forecast error counts; negative values absorb one-time seasonal shocks.

More negative Θ1\Theta_1Θ1​ produces a smoother seasonal forecast by dampening the influence of any single anomalous season. The forecast cone narrows beyond the first seasonal cycle (at h=2sh = 2sh=2s); the width at h=sh = sh=s, which sums the weights below lag sss, does not depend on Θ1\Theta_1Θ1​.

1
Explain: Size of random noise

How big one-step forecast errors are; sets overall noise and the width of forecast intervals.

Scaling σ\sigmaσ scales every drawn shock, so it widens or narrows all forecast intervals in proportion and scales the series' swings around 100, its seasonal pattern included: the toy has no seasonal pattern apart from the one the shocks build, so the seasonal amplitude goes from about 1.4 to 2.8 to 5.7 as σ\sigmaσ goes 0.5, 1, 2, while which months are high stays the same.

0
Explain: Drift

Constant in the doubly-differenced equation. For d=D=1d=D=1d=D=1, a non-zero drift implies a quadratic trend in levels - rarely appropriate for macro series, which is why this slider is deliberately narrow.

Non-zero drift tilts the forecast path up or down. Because both d=1d=1d=1 and D=1D=1D=1, even a small drift compounds into a noticeable curve over 24 forecast periods.

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
Aide
Politique de confidentialitéConditions d’utilisationÉthique et conformité

© 2026 Mark Jayson Nation LLC