Conditional projection methods for large-scale Bayesian VARs

arXiv:2607.29215 · econ.EM, stat.CO · Submitted 2026-07-31 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Conditional projection methods for large-scale Bayesian VARs".

Warren: Here's the gist of the paper. Conditional projection methods for large-scale Bayesian VARs develop fast methods for conditional forecasting and structural scenario analysis with high-dimensional Bayesian vector autoregressions (VARs).

Ruth: This framework enables fast and order-invariant equationby-equation estimation through a factor structure on the reduced-form errors.

Warren: Framework Description The general framework features a factor structure on the reduced-form errors, which enables fast and order-invariant equationby-equation estimation.

Ruth: This setup serves two purposes: first, the prior and factor structure enable regularization alongside quick and order-invariant estimation of VARs with a large cross section of n endogenous variables and long lag orders.

Title and authors: Iris: So to recap, the paper "Conditional projection methods for large-scale Bayesian VARs" focuses on developing a framework for fast estimation and simulation in high-dimensional VARs by using a factor structure on the reduced-form errors >

Ruth: Right. It’s not just about running the model; it’s about how they structure the problem so that even when you have huge systems, you can still get results quickly and consistently across all equations >

Warren: And they emphasize that this setup allows for a structural interpretation of those factors, meaning we can actually assign economic meaning to what those factors represent in the real economy >

Iris: They also mention that their framework builds on previous work from filtering-based methods or closed-form solutions for joint predictive distributions, but they extend it by adding these separate restrictions on observables and shocks >

Ruth: That extension is where the novelty lies. They are taking established ideas and layering in this more granular control over what you restrict for your scenarios >

Warren: It sounds like they’re building a tool that lets us move beyond just running one forecast to testing a wide variety of plausible economic futures based on different assumptions about the underlying shocks >

The paper's summary: Ruth: So the main summary is that they developed this framework to get conditional predictions with fast estimation and simulation algorithms for large Bayesian VARs >

Warren: And the big takeaway from their application is that you can use this method to compute counterfactual predictions for oil price scenarios in a specific context, like the two thousand twenty-six closure of the Strait of Hormuz > <ref:2607.29215#pg1,compute counterfactual predictions for oil price scenarios in>

Iris: They show that even with sign restrictions setting ten structural shocks for a US economy using thirty-three variables, they can compute these counterfactual predictions >

Ruth: And what’s really interesting is the empirical demonstration they provide: how the structural attribution of a scenario can drive vastly different counterfactual predictions >

Warren: They show that the same oil price path can lead to completely different outcomes, ranging from a mostly benign episode to pronounced stagflation depending on which structural and idiosyncratic shocks you allow in >

Iris: When they condition on observables, structural shocks, and idiosyncratic components together, they find that the required magnitude of a specific shock—like the structural oil price shock—increases by an order of magnitude >

The paper's improvements: Warren: Moving into how they improved things, the paper proposes a new algorithm for conditional projections that is fast and exact >

Ruth: And they’ve done something interesting with the cost analysis here. They say the cost is cubic only in the number of restrictions, while the size of the forecasted system enters only linearly >

Iris: That's a big deal for speed. It means you can get these projections much faster than earlier implementations when you are dealing with large systems that have many restrictions >

Warren: And they also show a comparison against four other samplers; their proposed algorithm dominates in empirically relevant cells, outperforming the joint samplers by factors of five to ten in bigger, sparsely restricted configurations >

Ruth: The precision sampler is better when things are more densely restricted, so it seems they’ve found a sweet spot for different types of problems based on how many restrictions you have >

Conclusion: Iris: So the final conclusion is that restricting the structural origins of a scenario is actually pertinent because the required shocks change significantly depending on which conditioning scheme you use >

Ruth: They show that when both margins are closed—meaning you restrict observables, idiosyncratic shocks, and structural shocks—the required magnitude of a specific shock, like oil price, increases by an order of magnitude >

Warren: That specific conditioning scheme produces a pronounced pattern of stagflation in their example >

Iris: But the authors also flag that getting certainty about the structural origin comes at the cost of much more diffuse counterfactual predictions >

Ruth: So, to wrap up, this paper on "Conditional projection methods for large-scale Bayesian VARs" shows how to handle huge systems using factor structures and fast samplers to analyze scenarios by separating restrictions across observables, shocks, and noise >

Warren: It’s a powerful tool for running counterfactuals that lets you see how sensitive your predictions are to what economic forces you assume are driving things >

Iris: And it gives us a concrete way to assess the trade-off between getting a very specific answer and having a broader range of possibilities >

University of Strathclyde · WU Vienna University of Economics and Business

econ.EM, stat.CO

Submitted: 2026-07-31

Updated: 2026-10-29

Comments: JEL: C11, C32, C53, E37, Q43; keywords: Structural scenario analysis, conditional forecasting, factor model

Code: https://github.com/mpfarrho/ssa-var-fm

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 91/100

The gist: Conditional projection methods for large-scale Bayesian VARs develop fast methods for conditional forecasting and structural scenario analysis with high-dimensional Bayesian vector autoregressions

Key concepts

Factor Structure on Reduced-Form Errors
This is a mathematical setup where the errors in a VAR model are broken down into a small number of common 'primitive' shocks (factors) and many independent, idiosyncratic shocks. This structure is key because it allows the estimation process to be done quickly and consistently, even when dealing with thousands of variables.
Equation-by-Equation Estimation
This refers to estimating each equation in a large VAR model individually rather than solving the entire system at once. The factor structure enables this estimation to be fast and order-invariant, meaning the speed of computation doesn't depend on how many lags are included in the model.
Structural Scenario Analysis (SSA)
This is a technique used to see how different economic scenarios arise based on which underlying shocks (structural vs. idiosyncratic) are allowed to happen. The paper shows that restricting the scenario to only structural shocks can lead to vastly different, and often more extreme, predictions compared to allowing all shock types.
Conditional Projection Algorithm
This is a fast sampling method proposed for calculating future forecasts based on current data. It uses auxiliary distributions and specific constraints to efficiently generate the final forecast distribution, making it computationally feasible for large systems.

Terminology

Summary

Conditional projection methods for large-scale Bayesian VARs develop fast methods for conditional forecasting and structural scenario analysis with high-dimensional Bayesian vector autoregressions (VARs). This framework enables fast and order-invariant equationby-equation estimation through a factor structure on the reduced-form errors.

Framework Description

The general framework features a factor structure on the reduced-form errors, which enables fast and order-invariant equationby-equation estimation. This setup serves two purposes: first, the prior and factor structure enable regularization alongside quick and order-invariant estimation of VARs with a large cross section of n endogenous variables and long lag orders. Second, following Korobilis (2022), a small number q (≪ n) of “primitive” shocks — the factors — drive the common dynamics, while cross-sectionally independent components capture idiosyncratic shocks.

Identification and Identification Assumptions

Identifying assumptions are crucial for structural interpretation, and this paper adopts a tradition to set-identify structural models via sensible sign and zero restrictions. The framework accommodates separate distributional restrictions on observables, structural shocks, and idiosyncratic components. Specifically, for the reduced-form error decomposition Σ = LL′ + S (as in Anderson et al., 1956), zero restrictions on L can achieve point identification at the cost of potentially arbitrary exclusion restrictions. Sign restrictions lead to set-identification, which is adopted here.

Conditional Projection Algorithm

The paper proposes a fast and exact sampler for conditional projections, with a cost that is cubic only in the number of restrictions while the size of the forecasted system enters linearly. The algorithm involves drawing from an auxiliary distribution N (0, Id) and N (0, Ω) independently to obtain restricted innovations ξ∗ ∼ N (µξ,Vξ). This allows for the construction of the final conditional projection yT +(1:H) R, r, omega ∼ N (µy,Vy).

Empirical Demonstration

An empirical demonstration shows that the structural attribution of a scenario can drive vastly different counterfactual predictions. The same oil price path is consistent with outcomes ranging from a mostly benign, mildly inflationary episode to pronounced stagflation, depending on which structural and idiosyncratic shocks are allowed to deliver it. For instance, under obs/idio/struc conditioning, the required magnitude of the structural oil price shock increases by an order of magnitude.

Computational Efficiency Comparison

The paper benchmarks Algorithm 1 against four competing samplers, showing that Algorithm 1 dominates in empirically relevant larger-system cells of the sparse-to-moderate density range. Specifically, in the biggest sparsely restricted configurations, Algorithm 1 outpaces the joint samplers by factors of five to ten. The precision sampler takes over in the most densely restricted settings.

Conclusion on Scenario Attribution

The paper concludes that restricting the structural origins of a scenario is pertinent, as the required shocks differ markedly across conditioning schemes. The results show that when both margins are closed (obs/idio/struc), the required magnitude of the structural oil price shock increases by an order of magnitude. This conditioning scheme produces a pronounced pattern of stagflation. The certainty about the structural origin comes at the price of much more diffuse counterfactual predictions.

Future Research Directions

Future research avenues include extensions of the model to include stochastic volatility and other forms of parameter time variation. The framework also lends itself to explicit manipulations of impulse responses to design policy counterfactuals. These methods are applicable with minor modifications when the model is Gaussian.

References

Adrian T, Giannone D, Luciani M, and West M (2026), “Risks and Uncertainty in Monetary Policy,” arXiv 2606.16708.

Anderson TW, Rubin H, et al. (1956), “Statistical inference in factor analysis,” in “Proceedings of the third Berkeley symposium on mathematical statistics and probability,” volume 5, 111–150.

Andersson MK, Palmqvist S, and Waggoner DF (2010), “Density-conditional forecasts in dynamic multivariate models,” Sveriges Riksbank Working Paper Series 247.

Antolin-Diaz J, Petrella I, and Rubio-Ram´ırez JF (2021), “Structural scenario analysis with SVARs,” Journal of Monetary Economics 117, 798–815.

Arias JE, Rubio-Ram´ırez JF, Rudolf D, and Shin M (2026), “Large SVARs,” Manuscript (unpublished).

Arias JE, Rubio-Ram´ırez JF, and Waggoner DF (2018), “Inference based on structural vector autoregressions identified with sign and zero restrictions: Theory and applications,” Econometrica 86(2), 685–720.

Banbura M, Bobeica E, and Mart ´ ´ınez Hernandez C ´ (2026), “What drives core inflation? The role of supply shocks,” ECB Working Paper 2875.

Banbura M, Giannone D, and Lenza M ´ (2015), “Conditional forecasts and scenario analysis with vector autoregressions for large cross-sections,” International Journal of Forecasting 31(3), 739–756.

Banbura M, Giannone D, and Reichlin L ´ (2010), “Large Bayesian vector auto regressions,” Journal of Applied Econometrics 25(1), 71–92.

Baumeister C (2025), “Discussion of ’Local Projections or VARs? A Primer for Macroeconomists’,” NBER Macroeconomics Annual forthcoming.

Baumeister C, and Hamilton JD (2015), “Sign restrictions, structural vector autoregressions, and useful prior information,” Econometrica 83(5), 1963–1999.

Berend L, and Pruser J ¨ (2025), “Large structural VARs with multiple linear shock and impact inequality restrictions,” arXiv 2505.19244.

Breitenlechner M, Geiger M, and Klein M (2024), “The fiscal channel of monetary policy,” WPs in Economics and Statistics 2024-07.

Caravello TE, McKay A, and Wolf CK (2025), “Evaluating Monetary Policy Counterfactuals:(When) Do We Need Structural Models?” NBER 32988.

Carriero A, Chan J, Clark TE, and Marcellino M (2024), “Addressing COVID-19 outliers in BVARs with stochastic volatility,” Review of Economics and Statistics 1–15.

Chan J, Eisenstat E, and Yu X (2022), “Large Bayesian VARs with factor stochastic volatility: Identification, order invariance and structural analysis,” arXiv 2207.03988.

Chan JC, Pettenuzzo D, Poon A, and Zhu D (2025), “Conditional Forecasts in Large Bayesian VARs with Multiple Equality and Inequality Constraints,” Journal of Economic Dynamics and Control 105061.

Crump RK, Eusepi S, Giannone D, Qian E, and Sbordonea A (2025), “A Large Bayesian VAR of the US Economy,” International Journal of Central Banking 21(2), 351–409.

Doan T, Litterman R, and Sims C (1984), “Forecasting and conditional projection using realistic prior distributions,” Econometric Reviews 3(1), 1–100.

Gambetti L, Korobilis D, Tsoukalas J, and Zanetti F (2023), “Agreed and disagreed uncertainty,” arXiv 2302.01621.

Huber F, and Feldkircher M (2019), “Adaptive shrinkage in Bayesian vector autoregressive models,” Journal of Business & Economic Statistics 37(1), 27–39.

Kastner G, and Huber F (2020), “Sparse Bayesian vector autoregressions in huge dimensions,” Journal of Forecasting 39(7), 1142–1165.

Makalic E, and Schmidt DF (2015), “A simple sampler for the horseshoe estimator,” IEEE Signal Processing Letters 23(1), 179–182.

Waggoner DF, and Zha T (1999), “Conditional forecasts in dynamic multivariate models,” Review of Economics and Statistics 81(4), 639–651.

Jarocinski M ´ (2010), “Conditional forecasts and uncertainty about forecast revisions in vector autoregressions,” Economics Letters 108(3), 257–259.

Improvements for AI systems

  1. Robust structural scenario analysis for high-dimensional systems: The improved system can perform structural scenario analysis (SSA) of Antolin-Diaz et al. (2021) by incorporating separate distributional restrictions on observables, structural shocks, and idiosyncratic components to define scenarios through distributional restrictions on observables, structural shocks and idiosyncratic components.

  2. Fast estimation for large Bayesian VARs: The system can achieve fast and order-invariant equation-by-equation estimation for VARs with a large cross section of endogenous variables by utilizing a factor structure on the reduced-form errors.

  3. Computationally efficient counterfactual forecasting: The improved system can compute counterfactual predictions for oil price scenarios in the context of the 2026 closure of the Strait of Hormuz with computational cost that is cubic only in the number of restrictions, while the dimension of the forecasted system enters linearly.

  4. Scenario-dependent macroeconomic insight: The improved AI can determine how the structural attribution of a scenario can drive vastly different counterfactual predictions, showing that "The same scenario is consistent with a benign, mildly inflationary episode when its structural origins are unrestricted, and with pronounced stagflation alongside substantially wider predictive intervals when it is attributed to the structural oil price shock alone."

  5. Precise uncertainty quantification: The system provides posterior medians and 68% credible sets for counterfactual forecasts under different conditioning schemes, allowing users to assess the impact of restricting shocks on predictive uncertainty, as seen in Figure 3.

Abstract

We develop methods for conditional forecasting and structural scenario analysis with high-dimensional Bayesian vector autoregressions (VARs). Our framework features a factor structure on the reduced-form errors, which enables fast and order-invariant equation-by-equation estimation; suitably identified factors admit a structural interpretation. We define scenarios through distributional restrictions on observables, structural shocks and idiosyncratic components. The computational cost of our proposed algorithm is cubic only in the number of restrictions, while the dimension of the forecasted system enters linearly. In our application with 37 variables for the US and ten structural shocks, identified by sign restrictions and internal instruments, we compute counterfactual predictions for oil price scenarios in the context of the 2026 closure of the Strait of Hormuz. Left to the model, the oil price path is read as an oil price shock accompanied by a demand boom; attributed to the structural oil price shock alone, a supply disruption, it implies higher inflation and slower growth.

Sources

Related papers