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

Forecast ensemble Model

Combination of several base forecasters - simple averages, weighted blends, or stacking - to reduce single-model risk.

OverviewGraphMathReferences

Why does averaging several mediocre forecasts almost always beat the single best individual forecast?

Background

Bates and Granger (1969) noticed something striking: a simple average of two airline-passenger forecasts beat each individual forecast. This was not a fluke. Four decades of subsequent evidence - surveyed by Timmermann (2006) and Genre, Kenny, Meyler, and Timmermann (2013) - confirms that forecast combination is one of the most reliable empirical regularities in economics. The "forecast combination puzzle" is that simple averages often beat sophisticated optimal weighting schemes, likely because the estimation error in optimal weights offsets their theoretical advantage in small samples.

The mechanism is variance reduction through diversification. If two forecasters make independent errors, their average has half the error variance of either one. In practice, forecast errors are correlated, but never perfectly so. As long as individual models capture partially distinct information, combining them cancels some errors. The mathematics is identical to portfolio diversification: the combined forecast is a portfolio of predictions, and the combination weights are portfolio weights that minimize forecast-error variance subject to the constraint that weights sum to one.

Modern practice goes beyond simple averaging. The ECB combines density forecasts from its Survey of Professional Forecasters using optimal linear pools. The Bank of England's Monetary Policy Committee informally combines outputs from multiple internal models (COMPASS, DSGE, SVARs) when producing its Inflation Report projections. The Federal Reserve Bank of Atlanta's GDPNow is itself a weighted combination of bridge equations, each using different indicator subsets. Academic researchers use Bayesian model averaging (BMA), where the weights are posterior model probabilities proportional to marginal likelihoods.

The field has moved from static to adaptive combination. Stock and Watson (2004) showed that the best individual forecasting model changes over time - a model that dominated in the 1990s may fail in the 2000s. Fixed weights estimated from a historical sample can assign high weight to a model whose advantage has evaporated. Time-varying weights that adapt to recent forecast performance - exponentially weighted moving averages of squared errors, Bayesian learning, or regime-switching weights - address this instability. Adaptive combinations sacrifice some statistical efficiency for robustness to structural change.

Connections to explore

Math ↗The structure and its conditions, step by step.References ↗Academic sources and guidance for using the model.

How the Parts Fit Together

y^m\hat{y}_my^​m​

Individual forecast

Point forecast from model mmm for the target variable yyy at forecast horizon hhh.

The inputs are MMM individual forecasts y^1,…,y^M\hat{y}_1, \ldots, \hat{y}_My^​1​,…,y^​M​ for a common target variable yyy at a common horizon hhh. Each forecast comes from a distinct model (VAR, DSGE, ARIMA, factor model, judgmental survey, etc.) or from the same model class with different specifications. The forecasts may be point forecasts (scalars) or density forecasts (full predictive distributions). Historical forecast errors are needed for weight estimation: a holdout sample of T0T_0T0​ periods where both the forecasts and the realized values are observed.

The combined forecast is y^c=∑m=1Mwmy^m\hat{y}_c = \sum_{m=1}^M w_m \hat{y}_my^​c​=∑m=1M​wm​y^​m​, where wm≥0w_m \geq 0wm​≥0 and ∑mwm=1\sum_m w_m = 1∑m​wm​=1. The weights wmw_mwm​ are chosen to minimize the forecast-error variance of the combination. Under the assumption that forecast errors have constant covariance Σe\Sigma_eΣe​ (M×MM \times MM×M matrix), the optimal weights are w∗=Σe−1ι/(ι′Σe−1ι)w^* = \Sigma_e^{-1} \iota / (\iota' \Sigma_e^{-1} \iota)w∗=Σe−1​ι/(ι′Σe−1​ι), where ι\iotaι is a vector of ones. This is the minimum-variance portfolio from finance. In practice, Σe\Sigma_eΣe​ must be estimated from the holdout sample, introducing estimation error that can make the estimated optimal weights worse than equal weights.

Weight estimation methods range from simple to complex. Equal weights (wm=1/Mw_m = 1/Mwm​=1/M) ignore the covariance structure entirely but avoid estimation error. Inverse-MSE weights set wm∝1/MSEmw_m \propto 1/\text{MSE}_mwm​∝1/MSEm​, using only individual model performance without cross-model covariance. Regression-based weights run the regression yt=β0+∑mβmy^m,t+εty_t = \beta_0 + \sum_m \beta_m \hat{y}_{m,t} + \varepsilon_tyt​=β0​+∑m​βm​y^​m,t​+εt​ and use β^m\hat{\beta}_mβ^​m​ as weights (Granger-Ramanathan, 1984), but this requires enough holdout data to estimate M+1M+1M+1 parameters reliably. Bayesian model averaging sets wmw_mwm​ proportional to the posterior model probability p(Mm∣data)p(\mathcal{M}_m | \text{data})p(Mm​∣data), computed from the marginal likelihood of each model. Time-varying weights use exponential decay: wm,t∝exp⁡(−α∑s=1t(ys−y^m,s)2)w_{m,t} \propto \exp(-\alpha \sum_{s=1}^{t} (y_s - \hat{y}_{m,s})^2)wm,t​∝exp(−α∑s=1t​(ys​−y^​m,s​)2) with a forgetting factor α\alphaα.

Assumptions

Testable

Stationary forecast-error distribution

The joint distribution of forecast errors (e1,…,eM)(e_1, \ldots, e_M)(e1​,…,eM​) is stationary over the holdout and forecast periods.

If violated:Weights estimated from historical errors are no longer optimal. Structural breaks, policy shifts, or model degradation change the error covariance over time.

Testable

Unbiased individual forecasts

Each individual forecast is unbiased: E[y−y^m]=0E[y - \hat{y}_m] = 0E[y−y^​m​]=0 for all mmm.

If violated:Biased forecasts transmit their bias to the combination unless the weights are constrained or a constant is included (Granger-Ramanathan regression). Mincer-Zarnowitz tests detect individual forecast bias.

Testable

Diversified information sets

Individual models use partially non-overlapping information, so their errors are imperfectly correlated.

If violated:If all models make the same errors (perfect correlation), combination provides zero variance reduction. The benefit of combining scales with the degree of error diversification.

Maintained

Sufficient holdout for weight estimation

The holdout sample T0T_0T0​ is large enough to estimate the covariance Σe\Sigma_eΣe​ reliably. For optimal weights, T0≫MT_0 \gg MT0​≫M.

If violated:Estimated optimal weights overfit the holdout sample. Sample covariance inversion amplifies estimation error. Equal weights become preferable when T0/MT_0 / MT0​/M is small.

Maintained

Correctly specified models (for BMA)

The true data-generating process is one of the MMM candidate models (or well-approximated by one).

If violated:BMA posterior concentrates on the single best-approximating model, missing the combination benefits that arise from averaging across structurally different models.

Maintained

Symmetric loss function

The loss function is quadratic (MSE). Over-prediction and under-prediction are equally costly.

If violated:Under asymmetric loss (e.g., underestimating inflation is worse than overestimating), MSE-optimal weights are suboptimal. Combination should target the asymmetric loss function directly.

Applications

The ECB's Survey of Professional Forecasters (SPF) collects density forecasts from approximately 60 institutions for euro-area GDP growth, inflation, and unemployment. The ECB staff combines these into a weighted density forecast, where the weights reflect each forecaster's past predictive accuracy. The combined density is tighter (lower entropy) than most individual densities, demonstrating the variance-reduction benefit. This combined forecast enters the Governing Council's deliberations alongside the ECB's own staff projections.

The Federal Reserve Bank of Philadelphia's Survey of Professional Forecasters publishes a consensus (median and mean) forecast that is itself a forecast combination. Research by Reifschneider and Tulip (2019) uses the historical errors of the SPF, the Fed, and other forecasters to gauge forecast uncertainty, finding Fed and private-sector accuracy broadly similar; Romer and Romer (2000), by contrast, documented an information advantage for the Fed's Greenbook (now Tealbook) forecasts. The combination of many diverse perspectives keeps the consensus competitive with a single well-resourced institution despite its confidential-data advantage.

Central banks that maintain multiple internal models combine them informally. The Bank of England's Monetary Policy Committee sees output from at least four models (COMPASS central forecast, DSGE variant, reduced-form VARs, judgmental overlays). The committee members weight these outputs subjectively, but the process is structured to ensure that no single model dominates. Research by Kapetanios, Mitchell, Price, and Fawcett (2015) formalized this by showing that density combination with time-varying weights improves upon any fixed-weight scheme for the Bank's inflation forecast.

Forecast combination fails when the model pool lacks diversity. Combining five VARs with slightly different lag lengths provides marginal benefit because their errors are nearly identical. The gains come from combining structurally different approaches: a statistical model (VAR) with a structural model (DSGE), a machine-learning method with a judgmental forecast, or a domestic model with one that emphasizes international linkages. Combination also fails when the number of models MMM is large relative to the holdout T0T_0T0​, because weight estimation becomes unreliable. In this regime, equal weights dominate, and the ensemble reduces to a simple average.

Extensions and open questions

Variants

  • Bayesian model averaging - weights proportional to posterior model probabilities
  • Exponentially weighted combination - recent forecast errors weighted more heavily, adapting to structural change
  • Stacked generalization - uses a meta-learner (regression, LASSO, neural network) to combine base model forecasts
  • Trimmed mean / median combination - robust to outlier forecasts by discarding extreme predictions
  • Density forecast combination - combines full predictive distributions rather than point forecasts alone

Open questions

  • Why simple averages outperform estimated optimal weights in most empirical settings (the forecast combination puzzle)
  • How to optimally combine density forecasts when individual predictive distributions are misspecified
  • Whether machine-learning-based meta-learners can reliably improve upon equal weighting for macro forecast combination

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