Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance
summary
The gist
This paper studies whether nonfundamentalness in small-scale VARs can be explained by omitted macroeconomic information and introduces a factor-filtering mixed causal-noncausal VARX framework to…
In short
The study tests if noncausal patterns in small-scale financial VAR models are due to genuine nonfundamentalness or simply omitted macroeconomic information. A factor-filtering mixed causal-noncausal VARX framework was introduced. Simulations and empirical application show that filtering common macroeconomic factors removes most noncausal components, suggesting missing information is the primary cause.
Key concepts
- Nonfundamentalness
- This refers to observed dynamics in financial time series that are not explained by underlying economic fundamentals. The paper investigates whether these patterns are real structural issues or artifacts of incomplete data being used in standard VAR models.
- Mixed Causal-Noncausal VARX
- This is a specialized model designed to handle time series that exhibit both causal (predictable) and noncausal (unpredictable) relationships. It allows researchers to explicitly test if the observed patterns are due to genuine structural dynamics or simply because common macroeconomic factors are missing from the data.
- Portmanteau Generalized Covariance (GCov) Estimator
- This is a statistical tool used to estimate parameters in complex VARX models. It works by minimizing a specific objective function based on nonlinear autocovariances of model errors, helping to correctly identify the causal and noncausal structures in the data.
- Factor Filtering
- This technique involves removing or 'residualizing' variables that share common macroeconomic factors. By filtering out these common influences, researchers can see if the remaining dynamics are purely causal or if noncausality disappears, indicating where the missing information lies.
Terminology used across episodes
This episode discusses
- Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance · Paper Radio
- Regularized Generalized Covariance (RGCov) Estimator
The paper
Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance · Read on arXiv
Maastricht University · University of Liege
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: "Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance".
Warren: Here's the gist of the paper. This paper studies whether nonfundamentalness in small-scale VARs can be explained by omitted macroeconomic information and introduces a factor-filtering mixed causal-noncausal VARX framework to assess this phenomenon.
Ruth: Pruning redundant reasoning tokens The central objective of the paper is to determine if noncausal dynamics observed in standard macro-finance VAR models reflect genuine nonfundamentalness or simply omitted information available to economic agents but unobserved by the econometrician.
Warren: This is achieved by introducing a factor-filtering mixed causal-noncausal VARX approach designed to account for common macroeconomic information.
Ruth: The authors assess the performance of this framework in simulated settings and show that the generalized covariance (GCov) estimator correctly recovers causal and noncausal dynamics when using several lags.
Title and authors: Ruth: So we’re looking at this paper titled "Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance." This one really gets to the core question: are those weird noncausal patterns we see in standard macroeconomic models actually real economic stuff, or is it just because we're leaving out some big, common data?
Warren: Yeah. The authors are trying to figure out if those dynamics mean something fundamental about the economy or if they're just echoes of macroeconomic information that agents know but the econometrician doesn't see. It’s about separating genuine nonfundamentalness from simple missing information in standard VAR models (<ref:2607.28131#pg2>).
Iris: They introduce a specific framework called a factor-filtering mixed causal-noncausal VARX approach to test this, which they assess in simulated settings to see if the generalized covariance estimator works correctly when you use several lags. This is the core methodology they’re testing (<ref:2607.28131#pg1>).
Ruth: And what’s really interesting is their empirical work with the Stock-Watson monetary policy VAR, which shows those noncausal components largely disappear once they filter out those common factors, and comparing the results to original data further removes the price puzzle (<ref:2607.28131#pg1>).
Warren: That’s a big deal if it holds up. It suggests that when you enrich your information set by filtering out common macroeconomic forces, you can get back to a cleaner causal structure, which directly impacts how we interpret monetary policy transmission (<ref:2607.28131#pg1>).
Iris: The way they build the model involves defining a mixed causal-noncausal model as a two sided moving average of order infinity, and they use this Portmanteau Generalized Covariance estimator to find parameters, which is essentially minimizing a function involving nonlinear autocovariances of the error terms (<ref:2607.28131#pg3>).
Ruth: And they test the consistency of that estimator using Monte Carlo results in a three dimensional VAR two framework, showing that the correct model selection rate goes up with the sample size, and when you include three factors, the estimator concentrates most of its mass at the true causal dimension seventy point one percent of replications (<ref:2607.28131#pg2>).
Warren: That’s a lot of evidence that this estimator is reliable in practice, especially when you’re dealing with small sample sizes where these issues usually pop up (<ref:2607.28131#pg4>).
Iris: The paper also shows that factor filtering progressively reduces the number of noncausal dimensions across all three policy specifications, and specifically when they use at least the first two macroeconomic factors, no noncausal dimension is detected in their Taylor-rule specifications (<ref:2607.28131#pg4>).
Ruth: So what does this mean for us as people who just look at these models? It suggests that those apparent noncausal dynamics in small information VARs might actually be caused by omitted common factors, not necessarily some deep, genuine economic weirdness (<ref:2607.28131#pg4>).
Warren: It changes how we think about the econometrician’s information set. It points to the idea that if you leave out relevant macroeconomic forces, those omissions can create a noncausal appearance when you shouldn't expect it (<ref:2607.28131#pg4>).
Title and authors: Iris: The paper also suggests that when relevant factors are omitted, the generalized covariance estimator might actually yield a non-zero dimension reflecting this informational mismatch, which is important for understanding where the error comes from (<ref:2607.28131#pg4>).
Ruth: It means we need to be careful about assuming structural shocks are always fundamental because that assumption might be too restrictive and not match the true structure of the data (<ref:2607.28131#pg4>).
Warren: And if we filter out those common factors, we see a clearer picture of how monetary policy shocks actually transmit, leading to a stronger disinflationary effect and reversing the unemployment response in their empirical application (<ref:2607.28131#pg1>).
Iris: The paper is also related to Beaudry et al. who also tackle this issue by developing a diagnostic for the severity of nonfundamentalness, relying on noncausal autoregressive models instead of invertible moving average representations (<ref:2607.28131#pg3>).
Ruth: So, to wrap up this segment on "Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance," the main implication is that we should be much more rigorous about accounting for common macroeconomic information before concluding that we’ve found genuine nonfundamental dynamics (<ref:2607.28131#pg1>).
Warren: We also see evidence from the simulations and empirical work showing that a richer information set, by filtering out those common factors, can restore the correct causal classification in these models (<ref:2607.28131#pg4>).
Iris: The paper suggests that when you use at least two macroeconomic factors for filtering, no noncausal dimension is detected in their model specifications, which strongly supports the idea that missing common information is a major driver here (<ref:2607.28131#pg4>).
Ruth: It’s about making sure our models reflect the actual constraints we are imposing on the data generating process, rather than just whatever noise or omission we happen to leave in there (<ref:2607.28131#pg4>).
Warren: So, while the paper shows that filtering helps clean up the results and removes what looks like noncausality, it also points out that if you omit relevant information, the generalized covariance estimator might still find a noncausal dimension reflecting that informational mismatch (<ref:2607.28131#pg4>).
Iris: The authors acknowledge a limitation in their approach by focusing on mixed causal-noncausal models and not fully exploring how to construct economically interpretable impulse response functions in these mixed setups (<ref:2607.28131#pg4>).
Ruth: That’s fair, because connecting the mathematical result back to a clear economic intuition is always the next step in this kind of work (<ref:2607.28131#pg4>).
Warren: We’ll leave that for later, but for now, we see that enriching our information set affects how we estimate policy shock transmission, leading to a more disinflationary response when filtering is applied (<ref:2607.28131#pg1>).
Iris: That's what the evidence shows about the impact of adding macroeconomic factors to the model, even if it doesn't solve every theoretical puzzle (<ref:2607.28131#pg4>).
Ruth: So, that’s where we’ll leave things for now on this paper discussing "Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance." We'll talk more later when we get a chance to look at how these tools could be used by the real markets.
The paper's summary: Iris: They are using a specific framework called a factor-filtering mixed causal-noncausal VARX approach. The goal is to test if the noncausality they see reflects genuine dynamics or just information that agents know but we don't observe.
Ruth: That makes sense, but how do they actually test it? What’s the technical setup?
Iris: They use the Portmanteau Generalized Covariance estimator. This method is used to estimate parameters in mixed causal-noncausal models. The authors show that this estimator correctly finds both causal and noncausal dynamics when you use several lags.
Warren: So, it’s a specific math tool they’ve designed for this kind of problem. What does that do for the results?
Iris: They also run Monte Carlo simulations to check the finite sample performance of that estimator in a VAR two framework. That tells us how reliable it is in real macroeconomic settings.
Ruth: And what did those simulations show about its reliability?
Iris: The results showed that the frequency with which you select the correct model goes up with the sample size, which means the estimator is consistent. Plus, when they include three factors, the estimator concentrates most of its mass at the true causal dimension in seventy point one percent of replications.
Warren: So, it seems like adding more information actually helps pinpoint where things are going in these models. What’s the practical application?
Ruth: The empirical application is to revisit a Stock-Watson model for inflation and unemployment under different policy rules. When they filter out variables using common factors from a database, those policy specifications disappear into purely causal representations.
Iris: And what happens when you look at the impulse response functions on these filtered systems?
Ruth: The impulse response functions computed on these factor-filtered systems resolve the price puzzle and actually lead to a stronger disinflationary effect from monetary policy. That’s a tangible result for anyone interested in how money affects the economy.
Warren: It sounds like adding that extra layer of information—the common factors—actually changes how we interpret policy shocks, leading to these clearer results.
Iris: Exactly. And they show that when you use at least the first two macroeconomic factors for filtering, no noncausal dimension is detected in their Taylor-rule specifications.
Ruth: So, what does this mean for someone just looking at these models? It suggests that those apparent noncausal dynamics in small information VARs might actually be caused by omitted common factors, not necessarily some deep economic weirdness.
Warren: It changes how we think about the econometrician’s information set. It points to the idea that if you leave out relevant macroeconomic forces, those omissions can create a noncausal appearance when you shouldn't expect it.
Iris: And they also note that when relevant factors are omitted, the generalized covariance estimator might still find a noncausal dimension reflecting this informational mismatch.
Ruth: It means we need to be careful about assuming structural shocks are always fundamental because that assumption might be too restrictive and not match the true structure of the data.
Warren: So, while the paper shows that filtering helps clean up the results and removes what looks like noncausality, it also points out that if you omit relevant information, the generalized covariance estimator might still find a noncausal dimension reflecting that informational mismatch.
Iris: They also admit a limitation in their approach by focusing on mixed causal-noncausal models and not fully exploring how to construct economically interpretable impulse response functions in these mixed setups.
Ruth: That’s fair, because connecting the mathematical result back to a clear economic intuition is always the next step in this kind of work.
Warren: But for now, we see that enriching our information set affects how we estimate policy shock transmission, leading to a more disinflationary response when filtering is applied.
Iris: That’s what the evidence shows about the impact of adding macroeconomic factors to the model, even if it doesn't solve every theoretical puzzle.
The paper's improvements: Ruth: So, we've seen how this paper uses factor filtering to test if noncausality is real or just missing data, and now we look at what they suggest we should do next to make these models better.
Warren: Right. What are the practical steps the authors are proposing? What’s the roadmap for using this kind of mixed causal-noncausal model in practice?
Iris: They suggest that when you use these factor filtering techniques, you can compute impulse response functions on purely causal, factor-filtered specifications. This lets you get standard recursive identification and gives us economically interpretable responses by resolving the price puzzle.
Ruth: That means we move from just detecting a problem to actually getting a usable economic picture from the model. How does that help people who only listen to the show?
Warren: It means that if you're running a VAR model, you don't just stop when you see weird noncausality; you actively try to enrich your data set with relevant macroeconomic factors first.
Iris: They also provide guidelines on how to apply that Portmanteau Generalized Covariance estimator in a VAR two context using standard starting values for optimization. It’s about making the estimation process more robust.
Ruth: So, it’s not just about filtering the data; it’s also about improving the math used to estimate those dynamics. What are they saying about that estimator itself?
Iris: The authors say the Portmanteau Generalized Covariance estimator is one step and it's consistent, asymptotically normal distributed, and semi-parametrically efficient. It gives us a good way to find parameters without getting bogged down in too many assumptions.
Warren: That sounds like a cost-saving measure for the quantitative side, because if you can estimate the right model efficiently, you save computation time and reduce error margins on your strategy.
Ruth: And they are also calling for more research on how to construct economically interpretable impulse response functions in these mixed causal-noncausal models. That’s where they admit their limitations.
Iris: They're flagging that connecting the math back to clear economic intuition is still a missing piece, especially when dealing with those complex mixed setups.
Warren: So, the immediate implication is that we should prioritize using these factor filtering methods if we suspect our small-scale VARs are suffering from omitted information.
Ruth: Exactly. The paper suggests that when you use at least the first two macroeconomic factors for filtering, no noncausal dimension is detected in their model specifications. That’s a strong finding for structuring future analysis.
Iris: It supports the hypothesis that if you're missing common information, that's what manifests as noncausality in small-scale systems rather than some deep fundamental issue.
Warren: So, the big shift here is moving from just looking at raw VAR results to actively enriching those results with filtered data before drawing any conclusions about economic structure.
Conclusion: Tom: So we're wrapping up this look at "Nonfundamentalness or missing information ? Evidence from causal-noncausal VARs in macro-finance." The main idea is that when you filter out common macroeconomic factors, those noncausal patterns often vanish, suggesting they were just due to omitted information.
Ruth: It’s a big conceptual shift. It makes us question whether some of the weird dynamics we see in these models are actually real economic structures or just artifacts of our incomplete data set.
Warren: From a trading perspective, it means that if you're running a model based on this framework, you need to be very careful about what factors you’ve included in your estimation process.
Iris: I just want to emphasize that the method they use, the Portmanteau Generalized Covariance estimator, is consistent and asymptotically normal distributed. That gives us confidence in the numbers they are presenting.
Ruth: And what does this mean for the world? It suggests that when we model complex systems like macro-finance, our assumption that structural shocks are always fundamental might be too restrictive.
Warren: If those assumptions aren't true, then the way we calculate risk and predict policy responses based on those models could be skewed by what we left out.
Iris: I just want to stress that while the framework is strong, they do acknowledge a limitation in their approach by not fully exploring how to get economically interpretable impulse response functions in these mixed setups.
Ruth: True, it's a model for identification, not necessarily a final economic interpretation tool yet. But the fact that filtering two factors removes noncausality is solid evidence for using richer data sets.
Warren: So, the practical implication is that we should be thinking about how to build more complete information sets into our quantitative strategies, not just how to tweak the math on a small model.
Iris: Exactly. It’s about acknowledging that the econometrician’s information set isn't always perfect, and we need to account for those missing pieces before concluding a phenomenon is fundamental.
Ruth: We’ve seen how enriching our information set affects policy shock transmission, leading to a stronger disinflationary response when filtering is applied. That’s a tangible result that matters for anyone tracking monetary policy.
Warren: It shows that the cost of running these models isn't just computational; it's also the cost of incomplete information driving your results.
Iris: I think it points toward a future where we might need to systematically test which macroeconomic factors are truly necessary to remove spurious noncausal signals.
Ruth: That’s the direction we’re heading with this research, looking for ways to identify those crucial factors that keep the model clean.
Warren: Next time, we'll be talking about how these ideas translate into actual trading strategies or risk management systems.
More episodes
- Daily Summary for 2026-11-19
- 2608.21274-Recommendation Quality and the Concentration of Consumption: Experimental Evidence from Netflix
- 2408.02757-A nonparametric test for diurnal variation in spot correlation processes
- 2306.16393-High-Dimensional Canonical Correlation Analysis
- 2605.17299-Geometric Brownian motion with intermittent entries and exits
- 2607.04753-Fooling Yourself: how narratives shape beliefs
- 2408.14671-Inference on Treatment Effects with Repeatedly Mismeasured High-Dimensional Controls
- 2201.06898-Difference-in-Differences Estimators for Treatments Continuously Distributed at Every Period
- 2501.15307-Influence Functions: A Unified Differentiation-Based Framework
- The day in conversation for 2026-11-19