Local Projections vs. VARs: Decoding the Economic Forecasting Face-Off
"Explore the surprising twists in the battle of Local Projections and Vector Autoregressions. Find out when to choose one over the other for superior economic predictions."
For years, economists have relied on sophisticated tools to predict the future of our economies. Among these, Local Projections (LPs) and Vector Autoregressions (VARs) have emerged as leading contenders, each with its own set of strengths and weaknesses. The debate over which method is superior has been ongoing, spurring countless studies and discussions. This article breaks down what recent research reveals about these two powerful tools, offering practical guidance for anyone involved in economic analysis and forecasting.
At their core, both LPs and VARs aim to estimate structural impulse responses, which are crucial for understanding how the economy reacts to different shocks. However, they approach this task from different angles. LPs directly project future outcomes onto current conditions, offering flexibility but potentially sacrificing precision. VARs, on the other hand, extrapolate long-term responses from short-term data, which can lead to smoother forecasts but may introduce bias. Understanding this bias-variance trade-off is key to choosing the right method.
Recent research, leveraging thousands of simulated data scenarios, sheds new light on this debate. By mimicking the complexities of the U.S. macroeconomic landscape, these simulations provide a rigorous testing ground for LPs and VARs, helping to clarify when each method is most effective. Whether you're a seasoned economist or just starting out, this guide will equip you with the knowledge to navigate the world of economic forecasting with confidence.
The Forecasting Challenge in Macroeconomics
Macroeconomists face persistent challenges in predicting how economies respond to shocks and policy changes, with forecasting accuracy remaining a central concern for researchers and policymakers alike. The choice of econometric methodology can materially affect the conclusions drawn about economic dynamics, making methodological comparisons a recurring focus of the discipline. While no single approach has emerged as universally superior, the debate between competing frameworks continues to shape how economic relationships are quantified and understood. The stakes of these methodological choices are substantial, given their influence on policy recommendations that affect millions of people.
LP and VAR Methods: Core Approaches and Known Weaknesses
Vector autoregressions (VARs) and local projections (LPs) represent two dominant frameworks for estimating impulse responses in applied macroeconomics, each with distinct tradeoffs. Jordà (2005) proposed local projections as a simpler alternative to VARs, estimating impulse responses separately at each horizon rather than iteratively, which reduces reliance on model specification assumptions. However, Herbst and Schenfельd (2024) demonstrated that LP point estimates can be severely biased when applied to the small samples commonly encountered in macroeconomic time series. Li, Mertens, and Stock (2022), in a large-scale comparison using thousands of macroeconomic variables, found that neither method consistently dominates, with each performing better under different data conditions. The MIT primer by Olea and colleagues (2025) underscores that applied macroeconomists must understand the statistical properties of both estimators to make informed methodological choices.
Evolving Econometric Traditions
The study of dynamic economic relationships has roots stretching back to the early-to-mid twentieth century, when economists began formalizing models of how variables interact over time. Structural econometric modeling was the dominant paradigm for decades, with researchers embedding economic theory directly into simultaneous equation systems. The emergence of reduced-form time series approaches, including vector autoregressions pioneered in the late 1970s and 1980s, marked a significant shift toward data-driven analysis. Local projections entered the literature more recently as a flexible alternative, broadening the toolkit available to macroeconomic researchers seeking to trace the effects of shocks.
Local Projections vs. VARs: Unveiling the Core Differences
Local Projections (LPs) and Vector Autoregressions (VARs) are powerful tools that economists use to understand how the economy reacts to different events, such as changes in interest rates or government spending. Both methods try to capture the 'impulse response,' which shows how key economic variables respond over time to an initial 'shock.' However, they go about this task in very different ways, leading to distinct strengths and weaknesses.
- Local Projections (LPs): Direct, flexible, but can be less precise.
- Vector Autoregressions (VARs): Model-based, smooth forecasts, but risk bias.
Ongoing Methodological Refinements
Recent years have seen a surge of interest in refining both LP and VAR methodologies, driven by growing computational capacity and expanding macroeconomic datasets. Researchers have explored modifications to basic local projections, including band-pass filtering and regularization techniques aimed at reducing the bias identified in smaller samples. Parallel work has focused on improving VAR estimation through better lag selection criteria and alternative identification strategies that address the perennial challenge of separating correlation from causation. Large-scale empirical exercises, such as those comparing thousands of variables, have begun to provide more systematic evidence about when each method performs well. The field remains active, with ongoing debates about optimal implementation practices for applied work.
Criticisms and Limitations
Despite their popularity, both LPs and VARs face significant criticisms that temper their practical utility. The small-sample bias documented in recent research raises concerns about the reliability of impulse response estimates based on limited macroeconomic time series data. VAR models, meanwhile, require strong assumptions about the structure of the economy and the number of lags to include, and misspecification in these dimensions can distort results. Neither method handles certain data features well, such as high persistence or structural breaks, which are common in macroeconomic applications. These limitations have prompted some researchers to advocate for hybrid approaches or to caution against overconfidence in any single framework.
Head-to-Head Comparison
Comparative analyses have generally found that the relative performance of LPs versus VARs depends heavily on the specific context and data characteristics at hand. VARs tend to offer greater efficiency when the model is correctly specified, extracting more information from a given dataset. LPs, by contrast, are more robust to misspecification because they do not impose a global model structure on the data. The tradeoff between these properties means that the choice between methods often comes down to the researcher's confidence in their model's structural assumptions and the size of the available sample. Applied researchers are increasingly encouraged to report results from both approaches as a robustness check.
The Future of Economic Forecasting: Combining the Best of Both Worlds
As our understanding of LPs and VARs continues to evolve, future research may explore hybrid approaches that combine the strengths of both methods. For example, researchers could use LPs to estimate short-term responses and VARs to extrapolate long-term trends, potentially achieving a more accurate and robust forecast. By carefully considering the bias-variance trade-off and tailoring their approach to the specific characteristics of the data, economists can continue to refine their forecasting tools and provide valuable insights into the ever-changing economic landscape.
Expert Perspectives on Methodological Choice
Leading macroeconomists have increasingly emphasized that the LP-versus-VAR debate should not be viewed as an either-or proposition. Rather, the two methods should be understood as complementary tools in the applied researcher's toolkit, each illuminating different aspects of dynamic economic relationships. Some experts have noted that the perceived simplicity of local projections can be deceptive, as the method still requires careful attention to issues like standard error construction and horizon selection. The consensus appears to be moving toward methodological pluralism, where the choice of estimator is guided by the specific research question rather than blanket preference for one approach over another.
Directions for Future Research
Several promising avenues for future research are emerging from the ongoing methodological debate. Machine learning techniques are beginning to be applied to the problem of impulse response estimation, potentially offering new ways to handle high-dimensional macroeconomic datasets. Researchers are also exploring how LPs and VARs can be adapted to incorporate real-time data and handle the kinds of irregular release schedules common in macroeconomic statistics. The development of better pre-testing procedures and model selection criteria tailored to the specific properties of each method remains an active area of investigation. As computational resources continue to expand, increasingly sophisticated simulation-based methods may help resolve some of the outstanding theoretical questions about these estimators.
Systemic Issues in Macroeconomic Modeling
The methodological debate sits within broader systemic challenges facing macroeconomic modeling, including the difficulty of capturing structural change in dynamic systems. Macroeconomic relationships are not stationary in the way that many econometric methods assume, and regimes can shift in ways that are difficult to anticipate. The challenge of model uncertainty—knowing which variables to include and how to specify their relationships—persists regardless of which estimation method is employed. These deeper issues suggest that progress in macroeconomic forecasting will require not just better estimation techniques but also more fundamental advances in how economists conceptualize economic dynamics.
From Theory to Policy
The practical significance of the LP-versus-VAR debate extends beyond academic methodology into the realm of real-world economic policy. Central banks and government agencies rely on impulse response analysis to evaluate the likely effects of monetary and fiscal policy decisions, making the choice of method a matter of practical consequence. Small differences in estimated impulse responses can translate into meaningfully different policy recommendations when the stakes involve billions of dollars in government spending or basis-point changes in interest rates. The human dimension of this technical debate underscores why methodological rigor in macroeconomics is not merely an academic exercise but a matter of genuine public importance.