Identification, estimation and inference in Panel Vector Autoregressions using external instruments
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: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "Identification, estimation and inference in Panel Vector Autoregressions using external instruments".
Warren: Here's the gist of the paper. This paper proposes an identification inspired from the SVAR-IV literature that uses external instruments to identify PVARs and discusses associated issues of identification, estimation, and inference<ref:2511.19372#pg1> It introduces a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a binary treatment<ref:2511.19372#pg1> Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses<ref:2511.19372#pg1> As an application, I instrument state-level military spending with the state’s share of national spending to estimate the dynamic fiscal multiplier<ref:2511.19372#pg1> I find multipliers above unity, with effects concentrated in the contemporaneous year and persisting into the following year<ref:2511.19372#pg1>.
Ruth: Causal Framework and Identification The paper establishes that causal effects can be identified by means of instrumental variables in the case of panel vector autoregressions<ref:2511.19372#pg1> By implementing the same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock and Watson (2018); Olea et al.
Warren: (2021) - defined as Proxy-SVAR or SVAR-IV- in PVARs it is possible to retrieve a causal estimand that I define as µ − LATE<ref:2511.19372#pg1> The definition of a µ − LATEj is given as E
Yj (w) − Yj (w′): for those units that comply with the assignment<ref:2511.19372#pg1>.
Ruth: The causal interpretation relies on three key assumptions: independence, exclusion and monotonicity<ref:2511.19372#pg1> Specifically, Assumption 4 requires the instrument to satisfy E
Wfi,t, Zi,t: ≠ 0<ref:2511.19372#pg1>.
Title and authors: Ruth: Let's talk about the title and who wrote this, "Identification, estimation and inference in Panel Vector Autoregressions using external instruments." It tells you exactly what's in there >
Warren: It signals that they're not just doing some standard regression; they are specifically looking at how to identify causal effects in panel vector autoregressions by using these external instruments >
Iris: They mention the core problem right away: standard panel vector autoregressions can have limited causal interpretation, sometimes only showing temporal related causality like Granger causality >
Ruth: And the authors say that exogeneity is hard to satisfy because of all the endogeneity among macroeconomic variables, and their proposed control type might not work for some units against others >
Warren: They suggest that instruments could be utilized when those previous cases fail, and this paper shows it's possible to identify causal effects using instrumental variables in panel vector autoregressions >
Iris: And they define this approach, which is sometimes called Proxy-SVAR or SVAR-IV in the paper, as a way to retrieve a causal estimand that they call the µ-LATE >
Ruth: So what does µ-LATE mean in plain terms? It's the average treatment effect for those specific units that comply with the assignment >
Warren: It’s an estimate of how much something changes when you switch from one assignment to another, specifically for those units that were actually assigned >
Iris: And the causal interpretation relies on three key assumptions: independence, exclusion, and monotonicity >
The paper's summary: Ruth: So let's talk about what they found in the summary of this paper. They propose using external instruments to identify panel vector autoregressions >
Warren: They introduce the concept of µ-LATE, that local average treatment effect, which arises when a continuous instrument targets a binary treatment >
Iris: The core idea is that under standard assumptions of independence, exclusion, and monotonicity, they show that externally instrumented panel vector autoregressions estimate this µ-LATE >
Ruth: They also show how to handle the estimation and inference issues associated with this method, which involves things like the Anderson Rubin statistics for confidence sets on impulse responses >
Warren: And they use Monte Carlo simulations to show that these confidence sets based on the Anderson Rubin statistics deliver reliable convergence for impulse responses >
Iris: As an application, they instrument state-level military spending with the state’s share of national spending to estimate the dynamic fiscal multiplier >
Ruth: And what did they find in that application? They found multipliers above one, and these effects are concentrated in the contemporaneous year and then persist into the following year >
The paper's improvements: Warren: Now for the improvements they suggest. They point out that their interpretation of µ-LATE relies on an underlying linearity assumption >
Ruth: So because of that, they suggest that non-parametric estimators could potentially alleviate this assumption if the data is large enough >
Iris: That makes sense from an econometric standpoint, because linearity is often a strong requirement for these types of identification methods when you're dealing with panel data >
Warren: They also flag the uncertainty around which statistics to use when dealing with multiple instruments, suggesting future researchers should properly discuss the unreliability of confidence sets in such cases >
Ruth: And they mention that even for large horizons in the impulse response function, the cumulative impulse response tends to have good coverage properties >
Conclusion: Ruth: So wrapping up on this paper, what's the main implication? They successfully applied these tools to estimate a dynamic regional fiscal multiplier for the United States >
Warren: The finding that multipliers can be above one in the second period suggests there might be longer term effects of fiscal expansions on GDP growth >
Iris: From an econometric perspective, it’s interesting that they found those multipliers persisted into the following year, which gives some weight to the results >
Ruth: They also suggest exploring non-parametric estimators if they have large datasets because of that linearity assumption issue >
Warren: And for those working with multiple instruments, they really stress the need to be careful about how reliable your confidence sets are when you use them >
econ.EM
Submitted: 2025-11-24
Updated: 2026-11-02
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 79/100
The gist: This paper proposes an identification inspired from the SVAR-IV literature that uses external instruments to identify PVARs and discusses associated issues of identification, estimation…
Key concepts
- $\mu$-LATE
- This is a specific causal target used when a continuous instrument is applied to a binary treatment. It measures the average treatment effect for those units that actually comply with the assignment of receiving the treatment versus not receiving it, providing a more precise causal interpretation than standard Granger causality.
- Instrumental Variables (IV) in PVARs
- This technique uses external variables, like state-level military spending instrumented by national spending shares, to isolate the causal effect of a primary variable. This helps identify causal effects in complex panel data structures where direct identification might be difficult.
- Anderson-Rubin Test Statistic
- This is an optimal test statistic used for inference when only one instrument is present in the analysis. It helps determine if the identified causal effect is statistically significant, offering a reliable way to draw conclusions from the estimation.
- Non-parametric Estimators
- These are estimation methods that do not rely on strict linearity assumptions about treatment effects. Researchers suggest exploring these if data is large enough, as they could potentially help overcome the limitations imposed by the parametric monotonicity assumption required for $\mu$-LATE identification.
Terminology
Summary
This paper proposes an identification inspired from the SVAR-IV literature that uses external instruments to identify PVARs and discusses associated issues of identification, estimation, and inference<ref:2511. p. 0 Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Under standard assumptions of independence, exclusion, and
monotonicity, I show that externally instrumented PVARs estimate the µ-LATE. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses. As an application, I instrument state-level military spending with the state’s share of national
spending to estimate the dynamic fiscal multiplier. As an application, I instrument state-level military spending with the state’s share of national
spending to estimate the dynamic fiscal multiplier. I find multipliers above unity,
with effects concentrated in the contemporaneous year and persisting into the following year.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Causal Framework and Identification
The paper establishes that causal effects can be identified by means of instrumental variables in the case of panel vector autoregressions<ref:2511.19372#pg1> By implementing the same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock and Watson (2018); Olea et al. p. 1 This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. (2021) - defined as Proxy-SVAR or SVAR-IV- in PVARs
it is possible to retrieve a causal estimand that I define as µ − LATE. Definition 1. A µ-LATEj
is the LATE of receiving assignment w versus assignment w
′
for those units that comply with the assignment
µ-LATEj = E
h
Yj (w) − Yj (w
′
)
W(z) = w
i
. 2 Assumptions
The assumptions required to give a causal interpretation to a LATE estimator are independence, exclusion and monotonicity. Here the assumptions are put on the residuals of the PVAR. (Monotonicity) For all z, z′ ∈ Zi,t, and all t ≥ 1 and i ≥ 1, it holds that
either Wfi,t(z) ≥ Wfi,t(z
′
) or Wfi,t(z) ≤ Wfi,t(z
′
). Notice that this type of monotonicity assumption needs to hold for every couple of
instrument assignments z, z′
.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Estimation Procedure
The estimation procedure involves a panel structural vector autoregression (PSVAR) where the focus is on the disturbances x˜i,t = (Wf′i,t, eY′j=1,i,t, eY′j=2,i,t,…)'<ref:2511.19372#pg1>. The model is represented as Rxi t = R(Im − Φ)µ i + RΦxi t−1 + Rx˜ i where x˜ i is the specific disturbance related to the original variable<ref:2511. p. 7 The focus of the following section will be on the disturbances
x˜i,t = (Wf′
i,t,
eY
′
j=1,i,t,
eY
′
j=2,i,t, . ,
eY
′
j=J,i,t)
′
,
where the tilde represents the specific disturbance related to the original variable. The reduced-form innovations xeit are related to the
SPVAR shocks ηt by an invertible matrix H:
xei,t = HΓηi,t = R
−1
ηi,t, ηi,t ∼ (0, IJ+1), diag(H) = 1,
where R−1 = HΓ, and Γ is a diagonal matrix with variance of the shocks in the diagonal
entries. The structural shocks ηi,t are mean zero with unit variance, serially and mutually
uncorrelated. The SPVAR is therefore under-identified as
there can be infinitely many solutions that satisfy the covariance restrictions. The IV procedure for the estimation of the structural matrix in SVARs generally
corresponds to either one or two separate steps, depending on whether the economist is
interested in a unit normalized shock or a standard shock11.Where the paper says it
Where the paper says it
Where the paper says it
Inference and Robustness
Inference can be carried by simply translating the conclusions of Olea et al. (2021) regarding VAR-IVs to their panel counterpart<ref:2511. p. 2 3
On the other hand, inference can be carried by simply translating the conclusions
of Olea et al. (2021) regarding VAR-IVs to their panel counterpart. This implies that
the optimal approach to inference is to make use of the Anderson Rubin test statistic
(henceforth defined as AR - see Fieller (1944); Anderson and Rubin (1949) for its definition) in the presence of one instrument, and, at the current state of the literature, the
conditional likelihood ratio test (henceforth defined as CLR - see Andrews et al. (2006)
for its definition) in the presence of multiple instruments. Finally, allow me to introduce some useful notation. A set of observations of variable
s for state i at time t, xs,it, is part of a larger matrix of variables xs, and x2:S indicates
4
In appendix 7. Under standard assumptions of independence, exclusion, and
monotonicity, I show that externally instrumented PVARs estimate the µ-LATE. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Application to Fiscal Multiplier
The paper applies these tools to estimate a regional fiscal multiplier for the United States using aggregate national military spending as an instrument for the innovations of regional military spending<ref:2511.19372#pg1> The empirical findings suggest that the dynamic regional fiscal multiplier may be above one in the second period, indicating some possibly longer term effects of fiscal expansions on GDP growth<ref:2511.19372#pg1>. The results are similar to those of the literature, suggesting a value of the fiscal multiplier of approximately ∼ 1. p. 3 My findings indicate
that the fiscal multiplier may be about ∼ 1. 7 on impact, a value not dissimilar from
the ones previously observed by the state-multiplier literature (Chodorow-Reich (2019)). My findings indicate
that the fiscal multiplier may be about ∼ 1. 7 on impact, a value not dissimilar from
the ones previously observed by the state-multiplier literature (Chodorow-Reich (2019)). ∗Email: raimondopala@gmail. com
1
arXiv:2511. ∗Email: raimondopala@gmail. com
1
arXiv:2511.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Future Directions
Future researchers are invited to develop two points: first, the µ − LATE interpretation relies on an underlying linearity assumption, and non-parametric estimators could alleviate this assumption if the data is sufficiently large<ref:2511.19372#pg1> Second, because the IV literature is still uncertain about which statistics to use when dealing with multiple instruments, future researchers should properly discuss the unreliability of confidence sets in such cases<ref:2511. p. 3 Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets. Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets.Where the paper says it
Where the paper says it
References
Anderson, T. W. and Rubin, H. p. 2 (2021) regarding VAR-IVs to their panel counterpart. This implies that
the optimal approach to inference is to make use of the Anderson Rubin test statistic
(henceforth defined as AR - see Fieller (1944); Anderson and Rubin (1949) for its definition) in the presence of one instrument, and, at the current state of the literature, the
conditional likelihood ratio test (henceforth defined as CLR - see Andrews et al. ∗Email: raimondopala@gmail. com
1
arXiv:2511. (2021) regarding VAR-IVs to their panel counterpart. This implies that
the optimal approach to inference is to make use of the Anderson Rubin test statistic
(henceforth defined as AR - see Fieller (1944); Anderson and Rubin (1949) for its definition) in the presence of one instrument, and, at the current state of the literature, the
conditional likelihood ratio test (henceforth defined as CLR - see Andrews et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. As such, the potential outcomes of the outcome variables can be represented as
follows:
Yj,i,t(w, z) = Yj,i,t((Wi,1:t−1, w, Wi,t+1:T )(Zi,1:t−1, z, Zi,t+1:T )). Such definition is similar to the one introduced by Rambachan and Shephard (2021) and
indicates that the potential outcome of the different outcome variables j depend on the
policy assignment w and on the instrumental variable z
5
. (2021) - defined as Proxy-SVAR or SVAR-IV- in PVARs
it is possible to retrieve a causal estimand that I define as µ − LATE. Normally, the
causal literature has a clear target in mind in the case of a dummy policy and a dummy
instrument (Angrist et al. (2021) regarding VAR-IVs to their panel counterpart. This implies that
the optimal approach to inference is to make use of the Anderson Rubin test statistic
(henceforth defined as AR - see Fieller (1944); Anderson and Rubin (1949) for its definition) in the presence of one instrument, and, at the current state of the literature, the
conditional likelihood ratio test (henceforth defined as CLR - see Andrews et al. However, when all the previous cases fail, instruments could be utilised to carry causal
claims. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. LATEs can be viewed as a special case
of principal stratification for which there are 4 categories of compliance status. In the
case of continuous instruments the interpretation becomes troublesome because there are
infinitely many possible strata (Antonelli et al. LATEs can be viewed as a special case
of principal stratification for which there are 4 categories of compliance status. In the
case of continuous instruments the interpretation becomes troublesome because there are
infinitely many possible strata (Antonelli et al. For this reason, I consider a PVAR that includes military state-level spending as
instrumented by the states fraction of total national military spending, with GDP growth
as the outcome variable. Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. For this reason, I consider a PVAR that includes military state-level spending as
instrumented by the states fraction of total national military spending, with GDP growth
as the outcome variable. Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. One natural contribution
is at the intersection of the causal inference and the time series fields. In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. (2022); Rambachan and Shephard (2021); Bojinov and
Shephard (2019)). In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. (2022); Rambachan and Shephard (2021); Bojinov and
Shephard (2019)). This paper contributes to several streams of the literature. One natural contribution
is at the intersection of the causal inference and the time series fields. Yet, such imposition would only refer
to a statistical relationship among the two, and would ignore the benefits of having a
7See, among many others, Stock and Watson (2018); Olea et al. (2021); Bruns and Keweloh (2024);
Brignone et al. Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Yet, such imposition would only refer
to a statistical relationship among the two, and would ignore the benefits of having a
7See, among many others, Stock and Watson (2018); Olea et al. (2021); Bruns and Keweloh (2024);
Brignone et al. Yet, such imposition would only refer
to a statistical relationship among the two, and would ignore the benefits of having a
7See, among many others, Stock and Watson (2018); Olea et al. (2021); Bruns and Keweloh (2024);
Brignone et al. My findings indicate
that the fiscal multiplier may be about ∼ 1. 7 on impact, a value not dissimilar from
the ones previously observed by the state-multiplier literature (Chodorow-Reich (2019)). As an application, I instrument state-level military spending with the state’s share of national
spending to estimate the dynamic fiscal multiplier. I find multipliers above unity,
with effects concentrated in the contemporaneous year and persisting into the following year. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses. As an application, I instrument state-level military spending with the state’s share of national
spending to estimate the dynamic fiscal multiplier. As an application, I instrument state-level military spending with the state’s share of national
spending to estimate the dynamic fiscal multiplier. I find multipliers above unity,
with effects concentrated in the contemporaneous year and persisting into the following year. (2021) regarding VAR-IVs to their panel counterpart. This implies that
the optimal approach to inference is to make use of the Anderson Rubin test statistic
(henceforth defined as AR - see Fieller (1944); Anderson and Rubin (1949) for its definition) in the presence of one instrument, and, at the current state of the literature, the
conditional likelihood ratio test (henceforth defined as CLR - see Andrews et al. ∗Email: raimondopala@gmail. com
1
arXiv:2511. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. Using
local level data can be advantageous for several reasons. First, Nakamura and Steinsson
(2014) argue that in a monetary union, the central bank (the Federal Reserve) cannot
raise interest rates in some states relative to others, and federal tax policy is common
across states in the union. intensifying (see Bartik (1991); Goldsmith-Pinkham et al. My findings indicate
that the fiscal multiplier may be about ∼ 1. Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. (1996)) or multi-valued
instruments (Heckman and Vytlacil (2001); Vytlacil (2002)), but rarely discussed how a LATE could
emerge from the comparison of of two predicted values of the outcome variables under a continuous
instrument case. However, their causal interpretation
is frequently limited. Sometimes they are described as identifying a temporally related
causality, “Granger causality”; sometimes, as discussed in Pala (2025), they may posses a
contemporaneous causal interpretation under either exogeneity assumptions, or a suitable
control type of assumption. In this special context, the defiers are (under positive monotonicity) units that do
not observe an increase (decrease) in their policy variable residuals when the instrument
increases (decreases). Moreover, because usually the residuals of a PVAR are assumed to
be continuous and normally distributed, the µ- LATE strongly depends on the instrument
1
In the words of Granger and Newbold (2014): “A better term might be temporally related, but
since it is such a simple term we shall continue to use it . This paper contributes to several streams of the literature. One natural contribution
is at the intersection of the causal inference and the time series fields. Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. (1996)) or multi-valued
instruments (Heckman and Vytlacil (2001); Vytlacil (2002)), but rarely discussed how a LATE could
emerge from the comparison of of two predicted values of the outcome variables under a continuous
instrument case. Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. (1996)) or multi-valued
instruments (Heckman and Vytlacil (2001); Vytlacil (2002)), but rarely discussed how a LATE could
emerge from the comparison of of two predicted values of the outcome variables under a continuous
instrument case. Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. In general, semi parametric or non parametric estimators tend to be preferred when
the data allows it. 3This point, and its many drawbacks, are developed in Koles´ar and Plagborg-Møller (2024). Panel vector autoregressions (PVAR) are one of the standard tools for the estimation of dynamic causal effects in macroeconomics. However, their causal interpretation
is frequently limited. One natural contribution
is at the intersection of the causal inference and the time series fields. In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. One natural contribution
is at the intersection of the causal inference and the time series fields. In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. (2006)
for its definition) in the presence of multiple instruments. I give a context to the proposed causal framework through an original application
in which I compute the dynamic fiscal military multiplier in the United States. Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets. Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets. Sometimes they are described as identifying a temporally related
causality, “Granger causality”; sometimes, as discussed in Pala (2025), they may posses a
contemporaneous causal interpretation under either exogeneity assumptions, or a suitable
control type of assumption. However, Granger causality is not contemporaneous causality1
; exogeneity is hard to satisfy due to the rich endogeneity among macroeconomic
variables (Nakamura and Steinsson (2018)); and the control type of approach proposed
by Pala (2025) may not be satisfied if some units cannot act as controls for some others. Sometimes they are described as identifying a temporally related
causality, “Granger causality”; sometimes, as discussed in Pala (2025), they may posses a
contemporaneous causal interpretation under either exogeneity assumptions, or a suitable
control type of assumption. However, Granger causality is not contemporaneous causality1
; exogeneity is hard to satisfy due to the rich endogeneity among macroeconomic
variables (Nakamura and Steinsson (2018)); and the control type of approach proposed
by Pala (2025) may not be satisfied if some units cannot act as controls for some others. Sometimes they are described as identifying a temporally related
causality, “Granger causality”; sometimes, as discussed in Pala (2025), they may posses a
contemporaneous causal interpretation under either exogeneity assumptions, or a suitable
control type of assumption. However, Granger causality is not contemporaneous causality1
; exogeneity is hard to satisfy due to the rich endogeneity among macroeconomic
variables (Nakamura and Steinsson (2018)); and the control type of approach proposed
by Pala (2025) may not be satisfied if some units cannot act as controls for some others. Sometimes they are described as identifying a temporally related
causality, “Granger causality”; sometimes, as discussed in Pala (2025), they may posses a
contemporaneous causal interpretation under either exogeneity assumptions, or a suitable
control type of assumption. However, Granger causality is not contemporaneous causality1
; exogeneity is hard to satisfy due to the rich endogeneity among macroeconomic
variables (Nakamura and Steinsson (2018)); and the control type of approach proposed
by Pala (2025) may not be satisfied if some units cannot act as controls for some others. First, Nakamura and Steinsson
(2014) argue that in a monetary union, the central bank (the Federal Reserve) cannot
raise interest rates in some states relative to others, and federal tax policy is common
across states in the union. This means that the open economy multiplier identifies the
measure of fiscal policy that does not depend on monetary movements. Panel vector autoregressions (PVAR) are one of the standard tools for the estimation of dynamic causal effects in macroeconomics. However, their causal interpretation
is frequently limited. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Identification, estimation and inference in Panel
Vector Autoregressions using external instruments
Raimondo Pala∗
November 25, 2025
Abstract
This paper proposes an identification inspired from the SVAR-IV literature that
uses external instruments to identify PVARs, and discusses associated issues of
identification, estimation, and inference. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Panel vector autoregressions (PVAR) are one of the standard tools for the estimation of dynamic causal effects in macroeconomics. However, their causal interpretation
is frequently limited. In this sense, it
complements Pala (2025) in the research agenda oriented to give a causal interpretation
to panel vector autoregressions and, more broadly, the literature that uses the Rubin
causal model to motivate the causal interpretation of time series models (Menchetti and
Bojinov (2022); Menchetti et al. (2022); Rambachan and Shephard (2021); Bojinov and
Shephard (2019)). This paper contributes to several streams of the literature. One natural contribution
is at the intersection of the causal inference and the time series fields. Second, the
advantage of using military spending is that the DD-350 military procurement forms
made available from the US Department of Defense provide a clean, direct, and localised
measure of government spending. Unfortunately, data of such quality, that starts from
such an early age - 1966 - is not available for any other form of spending. First, Nakamura and Steinsson
(2014) argue that in a monetary union, the central bank (the Federal Reserve) cannot
raise interest rates in some states relative to others, and federal tax policy is common
across states in the union. This means that the open economy multiplier identifies the
measure of fiscal policy that does not depend on monetary movements. This means that the open economy multiplier identifies the
measure of fiscal policy that does not depend on monetary movements. Second, the
advantage of using military spending is that the DD-350 military procurement forms
made available from the US Department of Defense provide a clean, direct, and localised
measure of government spending. However, when all the previous cases fail, instruments could be utilised to carry causal
claims. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. This paper shows that it is possible to identify causal effects by the means of instrumental variables in the case of panel vector autoregressions. By implementing the
same approach as Gertler and Karadi (2015); Mertens and Ravn (2013, 2014); Stock
and Watson (2018); Olea et al. Panel vector autoregressions (PVAR) are one of the standard tools for the estimation of dynamic causal effects in macroeconomics. However, their causal interpretation
is frequently limited. with the following first and second stages
Wfi,t = δZi,t + ηi,t first stage
eYj,i,t = βjWfi,t + ϵi,t second stage(s)
Then, if the economist is interested in a unit normalized shock, the IV estimator simply
becomes β
IV
j = (ρj/δ), where ρj comes from the regression eYj,i,t = ρjZi,t + νi,t, R1,1 = δ
,and R1,j = β
IV
j
. In the case in which the economist is interested in a standardized shock,
instead, the IV estimator becomes β
IV
j = cj (ρj/δ). Such IV approach is generally defined as Bartik instrument and
is becoming increasingly popular in social sciences, where the use of local-level data is
2
In fact, most of the IV literature has focused on either binary (Angrist et al. (1996)) or multi-valued
instruments (Heckman and Vytlacil (2001); Vytlacil (2002)), but rarely discussed how a LATE could
emerge from the comparison of of two predicted values of the outcome variables under a continuous
instrument case. I introduce a form of local average treatment effect - the µ-LATE - which arises when a continuous instrument targets a
binary treatment. Under standard assumptions of independence, exclusion, and
monotonicity, I show that externally instrumented PVARs estimate the µ-LATE. 7 on impact, a value not dissimilar from
the ones previously observed by the state-multiplier literature (Chodorow-Reich (2019)). The advantage of using PVARs over simple panel linear regressions becomes clear when
considering the usefulness of Impulse Response Functions.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Appendix Details
Appendix 7.3 demonstrates the good small sample properties of the AR confidence sets calibrating a simulation on the basis of the dataset from the application<ref:2511. p. 3 Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets. Under standard assumptions of independence, exclusion, and
monotonicity, I show that externally instrumented PVARs estimate the µ-LATE. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses. Third, it
indicates a potentially interesting recipe for the interpretation of any estimator in the case
of continuous instruments that is different from the current literature. Fourth, it extends
the common knowledge of the good coverage properties of the Anderson-Rubin statistics
for the impulse response functions generated by a PVAR-IV, indicating a venue for the
construction of reliable confidence sets. Under standard assumptions of independence, exclusion, and
monotonicity, I show that externally instrumented PVARs estimate the µ-LATE. Monte Carlo simulations illustrate that confidence sets based on the AndersonRubin statistics deliver reliable convergence for impulse responses.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Table Data
Table 4 and Table 5 present MAIC, BBIC, and MHIQ tests for different lag specifications<ref:2511.19372#pg1> Table 6 shows the autoregressive coefficients of the PVAR estimated using growth levels with respect to two years before and two lags<ref:2511.19372#pg1>. The results in Table 3 show the autoregressive coefficients of the PVAR<ref:2511.19372#pg1>. Table 2 reports the result of the regression of state spending innovations on total spending, with standard confidence intervals and first stage F-statistic<ref:2511. p. 9 with the following first and second stages
Wfi,t = δZi,t + ηi,t first stage
eYj,i,t = βjWfi,t + ϵi,t second stage(s)
Then, if the economist is interested in a unit normalized shock, the IV estimator simply
becomes β
IV
j = (ρj/δ), where ρj comes from the regression eYj,i,t = ρjZi,t + νi,t, R1,1 = δ
,and R1,j = β
IV
j
. In the case in which the economist is interested in a standardized shock,
instead, the IV estimator becomes β
IV
j = cj (ρj/δ).Where the paper says it
Conclusion
The paper concludes that the dynamic regional fiscal multiplier may be above one in the second period, indicating some possibly longer term effects of fiscal expansions on GDP growth<ref:2511.19372#pg1>. The authors suggest that non-parametric estimators could be explored if the data is sufficiently large to alleviate the linearity assumption<ref:2511. p. 2 (1996)) or multi-valued
instruments (Heckman and Vytlacil (2001); Vytlacil (2002)), but rarely discussed how a LATE could
emerge from the comparison of of two predicted values of the outcome variables under a continuous
instrument case. In general, semi parametric or non parametric estimators tend to be preferred when
the data allows it. This choice stems
from the higher likelihood of researchers to come up with identification strategies that involve a clear
one insturment, one instrumented, many outcomes scenarios. It can be extended to multiple instruments
and multiple instrumented variables. (2006)
for its definition) in the presence of multiple instruments. I give a context to the proposed causal framework through an original application
in which I compute the dynamic fiscal military multiplier in the United States.Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improve Causal Interpretation of Panel Vector Autoregressions by defining µ-LATE as
the LATE of receiving assignment w versus assignment w' for those units that comply with the assignment
to provide a clearer causal target than standard Granger causality. -
Enhance Dynamic Fiscal Multiplier Estimation by using state-level military spending instrumented by the state’s share of national military spending to estimate the dynamic fiscal multiplier, finding effects
concentrated in the contemporaneous year and persisting into the following year.
-
Strengthen Inference Capabilities by recommending a two-step procedure involving first stage F-statistics and second stage Anderson Rubin statistics or bootstrap methods, as this approach
may be sub optimal compared to the Anderson-Rubin statistic approach.
-
Improve Impulse Response Analysis by using the AR statistic for confidence sets on impulse response functions, which is shown to deliver
good coverage properties
even in small sample scenarios. -
Develop Robust Estimation Methods by exploring non-parametric estimators for continuous instruments, as they
could potentially alleviate the linearity assumption
inherent in the parametric monotonicity assumption required for µ-LATE identification.
Abstract
This paper proposes an identification inspired by the SVAR-IV literature that uses external instruments to identify panel vector autoregressions (PVARs), and discusses associated issues of identification, estimation, and inference. I introduce a form of local average treatment effect, the μ-LATE, which arises when a continuous instrument targets a policy variable with linear effects, and which reduces to the local average treatment effect when the policy and the instrument are binary. Under standard assumptions of independence, exclusion, and monotonicity, I show that externally instrumented PVARs estimate the μ-LATE. I construct confidence sets for the impulse responses based on the Anderson-Rubin statistic and, with several instruments, on conditional likelihood ratio tests, which are robust to weak instruments and to common shocks across units. I apply the method to the effect of stock market wealth on local labor markets across US counties.
Sources
- Principal stratification with continuous treatments and continuous post-treatment variables
- Robust Impulse Responses using External Instruments: the Role of Information
- Dynamic Causal Effects in a Nonlinear World: the Good, the Bad, and the Ugly
- The causal interpretation of panel vector autoregressions
Related papers
- Stationarity with Occasionally Binding Constraints
- Backtesting High-Dimensional Expected Shortfall
- Higher-Order Neyman Orthogonality in Moment-Condition Models
- Online Bayesian Inference for a Dirichlet Process Mixture of Binary Logit Models
- Debiased Conditional Independence Tests for Causal Effects
- Sharp Testable Implications of Encouragement Designs