Time Series Econometrics
Time series econometrics is the branch of econometrics that deals with data observed in sequence over time—daily stock returns, quarterly GDP, monthly inflation, hourly electricity demand.
11 frameworks, from 1920 to today · Econometrics · Economics
- Classical Business Cycle Analysis 1920-1960
- Stationary process 1930-Present
- State-Space Models and the Kalman Filter 1960-Present
- Box–Jenkins method 1970-Present
- Unit Root and Spurious Regression Analysis 1974-Present
- Vector autoregression 1980-Present
- Autoregressive conditional heteroskedasticity 1982-Present
- Cointegration 1987-Present
- Bayesian Time Series Analysis 1990-Present
- Local Projections 2005-Present
- Machine Learning for Time Series Forecasting 2010-Present
Overview
Time series econometrics is the branch of econometrics that deals with data observed in sequence over time—daily stock returns, quarterly GDP, monthly inflation, hourly electricity demand1. What distinguishes it from other kinds of econometrics is that the order of the observations carries information1. Today's value of a series typically depends on yesterday's, and the shocks that hit an economy persist, fade, or compound over time. The field's central tasks are to describe that dependence, to forecast what comes next, to test whether apparent relationships between series are real, and to identify what caused what. Its history is a story of successive crises: each generation's standard tools broke down on some feature of real data, and the repair became the next generation's toolkit.
Cycles, Decomposition, and the Prewar Statistical Tradition
Nineteenth- and early twentieth-century economists studied business cycles by eye and by index number. The turning point came with the idea that an economic series could be split into components—a long-run trend, a periodic business cycle, a seasonal pattern, and irregular noise—and that each could be estimated separately. The classical business cycle decomposition, developed in the 1920s and 1930s by figures such as Warren Persons and the National Bureau of Economic Research under Wesley Mitchell and Arthur Burns, formalized this: series were smoothed, peaks and troughs dated, and cycles measured by their duration and amplitude. The NBER's dating of expansions and recessions descends directly from this work.
A parallel statistical tradition grew out of the study of random processes. The Russian mathematician Andrey Kolmogorov and the British statistician Gilbert Walker, working on weather and later on the Indian monsoon, developed the mathematics of stationary stochastic processes—sequences whose statistical properties (mean, variance, autocorrelation) do not change over time. Walker's work in the 1920s and 1930s on autoregressive models, in which a series is regressed on its own past values, gave the field its basic vocabulary. The Swedish economist Herman Wold's 1938 thesis on stationary processes and his later work established that a broad class of stationary series can be represented as a moving average of past random shocks. These were the theoretical foundations on which everything later rested.
The Box–Jenkins Revolution
For decades, forecasting was an art of judgment and large structural models. That changed with the 1970 publication of Time Series Analysis: Forecasting and Control by the statisticians George Box and Gwilym Jenkins2. Their ARIMA methodology—autoregressive integrated moving average—gave practitioners a systematic procedure: identify the order of the model from the data's autocorrelation and partial autocorrelation functions, estimate the parameters, check the residuals, and forecast34. The "integrated" part handled nonstationarity by differencing the series until it was stationary5.
Box–Jenkins was a shock to econometrics because it was deliberately atheoretical. It forecast from the series' own history, without an economic model. In the 1970s, when large Keynesian macroeconometric models performed poorly during stagflation, this unpretentious approach often beat them in forecasting competitions. The lesson—that a well-specified statistical model of the data's own dynamics can be hard to beat—has never been forgotten.
Spurious Regression and the Unit Root Problem
The next crisis was statistical. In 1974, Clive Granger and Paul Newbold published a startling simulation: if you regress one random walk on another independent random walk, the regression will frequently show a strong, "significant" relationship that does not exist67. A random walk is a series whose changes are unpredictable noise—a stock price is the classic example—and it is nonstationary: its variance grows without bound. The finding, known as spurious regression, meant that standard regression inference was invalid for many macroeconomic series.
The diagnosis crystallized around the concept of the unit root: a series whose autoregressive coefficient equals one, so that shocks never die out89. Testing for unit roots became a major industry8. David Dickey and Wayne Fuller developed the Dickey–Fuller test in the late 1970s, and later augmented versions and the Phillips–Perron test (1988) followed. The practical message was that one must first determine whether a series is stationary—mean-reverting—before regressing it on anything6.
Cointegration and Error Correction
Granger and Robert Engle's 1987 paper resolved the apparent dead end6. Two nonstationary series can wander individually yet stay tethered together—like a drunk and her dog, in the famous analogy6. If a linear combination of them is stationary, the series are cointegrated, and regressing one on the other is not spurious but estimates a genuine long-run relationship6. Moreover, the tether implies an error correction mechanism: when the series drift apart, forces pull them back7. Engle and Granger's representation theorem showed that cointegration and error correction are two faces of the same phenomenon. Their 1987 paper won the 2003 Nobel Prize.
The idea transformed empirical macroeconomics. Consumption and income, money demand and interest rates, exchange rates and prices—these could now be modeled as long-run equilibria with short-run adjustment7. Søren Johansen's 1988 and 1991 work extended the framework to systems of many variables, providing maximum-likelihood tests for the number of cointegrating relationships. Error correction models became the standard way to combine long-run theory with short-run dynamics.
Vector Autoregressions and the Identification Problem
In 1980, Christopher Sims published a critique of the large structural models then dominant: their identifying assumptions—which variables were exogenous, which equations were simultaneous—were "incredible." His alternative, the vector autoregression (VAR), treated every variable as endogenous and modeled each as a function of the past values of all the others. VARs were atheoretical but honest, and they became the workhorse of empirical macroeconomics.
The catch was that a VAR's reduced-form residuals are correlated across equations, so one cannot tell which shock moved which variable. Sims's original solution imposed a recursive ordering—a Cholesky decomposition—in which some variables respond to shocks only with a lag. Later work, notably by Olivier Blanchard and Danny Quah in 1989, imposed long-run restrictions instead. The structural VAR (SVAR) literature that grew from these debates is essentially a continuing argument about what economic theory licenses one to assume in order to interpret the data's correlations as causal shocks.
Volatility Modeling
Financial series posed a different problem. Asset returns show little autocorrelation in their levels but strong clustering in their variance: turbulent periods follow turbulent periods1011. Robert Engle's 1982 paper introduced the autoregressive conditional heteroskedasticity (ARCH) model, in which the conditional variance of a series depends on the squares of past shocks1011. Tim Bollerslev's 1986 generalization, GARCH, added lagged variances and became the standard tool for modeling and forecasting financial volatility. The framework underlies option pricing, risk management, and the value-at-risk measures used across finance. Engle shared the 2003 Nobel with Granger.
State-Space Models and the Kalman Filter
A separate lineage runs through engineering. The Kalman filter, developed by Rudolf Kalman in 1960 for guidance and control, provides a recursive way to update estimates of unobserved quantities as new data arrive. In econometrics, state-space models represent a series as driven by hidden components—a slowly moving trend, a seasonal factor, a time-varying parameter—and the filter extracts estimates of those components from noisy observations12. The approach, championed in econometrics by Andrew Harvey and others from the 1980s, unified many earlier ad hoc smoothing methods and made it possible to estimate models with unobserved variables, including time-varying coefficients and dynamic factor models of many series.
Bayesian Time Series
Bayesian methods entered time series through the same door as state-space models: both treat parameters as things to be learned from data rather than fixed. The Bayesian approach—treating every unknown as a random variable with a prior distribution, updated to a posterior by the data—was computationally impractical for large models until the development of Markov chain Monte Carlo methods in the late 1980s and 1990s. Since then, Bayesian VARs with shrinkage priors have become a leading tool for macroeconomic forecasting, and Bayesian estimation of dynamic stochastic general equilibrium models is central to modern macro. The Bayesian framework's appeal in time series is natural: forecasting is inherently a statement of belief about the future, and the Bayesian formalism makes that explicit.
Local Projections
In 2005, Òscar Jordà proposed a simple alternative to VARs for tracing the effects of a shock over time. Instead of estimating a VAR and simulating its impulse responses, one could run a separate regression for each horizon—regressing the outcome at horizon h directly on the shock and controls. These local projections are robust to misspecification of the underlying dynamics, easy to implement, and flexible about nonlinearity. They have become widely used in empirical macroeconomics, especially for estimating the effects of monetary and fiscal policy.
Machine Learning and High-Dimensional Forecasting
The most recent wave brings the tools of machine learning into time series forecasting1. Modern datasets—scanner data, satellite imagery, text, hundreds of macroeconomic indicators—have more predictors than observations, a setting where classical regression fails. Regularization methods such as the lasso, tree-based ensembles, and neural networks can handle this. The results have been mixed but instructive: simple benchmarks remain hard to beat in many forecasting tasks, and the gains from machine learning tend to appear where the data are rich and the relationships nonlinear. The field is now working out when and why these methods help, and how to combine them with the structural insights of the earlier traditions.
The Present
Today's time series econometrics is a layered toolkit. Unit root and cointegration analysis remains the first step for any long macroeconomic series6. VARs and their structural variants, alongside local projections, are the standard instruments for policy analysis. GARCH-family models dominate financial volatility work. State-space methods and Bayesian estimation underpin forecasting systems at central banks. Machine learning is being absorbed as an additional set of estimators rather than a replacement for the field's core concerns. The open questions are familiar in shape: how to identify causal effects from observational time series, how to forecast in the presence of structural breaks and regime changes, and how to reconcile the atheoretical predictive power of statistical models with the interpretability that economic questions demand.
References
- Time series (introduction, Other modeling approaches), Wikipedia
- Autoregressive moving-average model (introduction, History and interpretations), Wikipedia
- Box–Jenkins method, Wikipedia
- Partial autocorrelation function, Wikipedia
- Autoregressive integrated moving average, Wikipedia
- Cointegration (History, introduction), Wikipedia
- Error correction model (History, introduction), Wikipedia
- Unit root, Wikipedia
- Stationary process, Wikipedia
- Autoregressive conditional heteroskedasticity, Wikipedia
- Volatility clustering, Wikipedia
- State-space representation, Wikipedia
The frameworks of Time Series Econometrics
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1920-1960
- 1930-Present
- 1960-Present
- 1970-Present
- 1974-Present
- 1980-Present
- 1982-Present
Autoregressive conditional heteroskedasticity
About Autoregressive conditional heteroskedasticity
Classical time series models, from the Box-Jenkins ARIMA tradition to Vector Autoregressions, share a foundational assumption: the variance of the error term is constant over time. This assumption, known as homoskedasticity, is convenient but often false for economic and financial data. Asset returns, exchange rates, and inflation rates exhibit periods of calm followed by periods of turbulence. A model that assumes a constant variance will systematically underestimate uncertainty during volatile episodes and overestimate it during quiet ones, leading to poor risk assessments and unreliable forecast intervals. The ARCH and GARCH framework was developed precisely to address this failure.The distinctive organizing idea of this framework is that volatility is not a nuisance parameter but a variable to be modeled. Instead of treating the variance as a fixed constant, the framework treats the conditional variance—the variance of the error term given the information available at time —as a function of the past. The core insight is that large shocks tend to be followed by more large shocks, and small shocks by more small shocks. This phenomenon, known as volatility clustering, is the empirical signature that the framework captures. The framework does not attempt to explain why volatility changes; it models the serial dependence in the variance itself.The intellectual history begins with Robert Engle’s 1982 paper, which introduced the Autoregressive Conditional Heteroskedasticity (ARCH) model. Engle’s insight was to make the conditional variance a linear function of the squares of past error terms. In an model, the variance at time is \(\sigma_t^2 = \alpha_0 + \alpha_1 \varepsilon_{t-1}^2 + \dots + \alpha_q \varepsilon_{t-q}^2\). This specification allows the model to capture volatility clustering: a large past error increases today’s variance, making another large error more likely. The model was a conceptual breakthrough because it showed that heteroskedasticity—non-constant variance—could be modeled parametrically rather than merely corrected for.A practical limitation of the ARCH model is that it often requires a high order to capture the persistence of volatility shocks. A natural extension, introduced by Tim Bollerslev in 1986, is the Generalized ARCH (GARCH) model. In a model, the conditional variance depends not only on past squared errors but also on its own past values: \(\sigma_t^2 = \alpha_0 + \sum_{i=1}^q \alpha_i \varepsilon_{t-i}^2 + \sum_{j=1}^p \beta_j \sigma_{t-j}^2\). This addition of autoregressive terms for the variance allows a parsimonious representation of long memory in volatility. The most widely used specification, , uses only one lag of each and is often sufficient for financial data. The GARCH model became the workhorse of the framework, and its success is reflected in Engle’s 2003 Nobel Prize in Economic Sciences.To understand the framework’s place, it is useful to contrast it with its neighbors. The Box-Jenkins ARIMA framework models the conditional mean of a series as a function of its past, assuming a constant variance. ARCH and GARCH models complement this by modeling the conditional variance; in practice, one often fits an ARIMA model for the mean and a GARCH model for the variance of the residuals. Vector Autoregressions extend mean modeling to multiple series but retain the homoskedasticity assumption. The successor framework, Structural Break and Regime-Switching Models, takes a different approach to instability: it assumes that the data-generating process switches between distinct regimes, each with its own parameters. GARCH models, by contrast, assume a single continuous process with smoothly evolving variance. The distinction is subtle but important: GARCH captures volatility that changes gradually and predictably from its own past, while regime-switching models capture abrupt changes in the entire data-generating process.The essential concepts of the framework can be learned in a logical sequence. First, one must understand conditional heteroskedasticity: the idea that the variance of the error term, conditional on past information, is not constant. Second, one must grasp volatility clustering as the empirical motivation. Third, the ARCH model provides the first parametric specification. Fourth, the GARCH model generalizes it by adding autoregressive terms for the variance. From there, one encounters a family of extensions that address specific limitations: EGARCH (exponential GARCH) allows for asymmetric responses to positive and negative shocks, capturing the leverage effect where bad news increases volatility more than good news; GJR-GARCH also models asymmetry; and IGARCH (integrated GARCH) handles cases where the variance process has a unit root, implying that shocks to volatility are permanent. For most practical purposes, however, the model remains the default starting point.The continuing significance of the framework lies in its application to risk management, option pricing, and financial regulation. Value-at-Risk (VaR) calculations, which estimate the maximum loss over a given horizon at a given confidence level, are dramatically improved by GARCH-based forecasts of volatility. The framework also underpins the concept of volatility forecasting: because the conditional variance is a function of the past, it can be forecasted one step ahead, and multi-step forecasts can be generated recursively. This predictive ability is the framework’s practical payoff. While more sophisticated models—stochastic volatility models, realized volatility measures, and machine learning approaches—have emerged, GARCH models remain the benchmark against which new methods are judged, and they are a standard component of any econometrics curriculum.For the reader, the prerequisites are modest but real. A working knowledge of linear regression and the concept of maximum likelihood estimation is essential, as GARCH models are estimated by maximum likelihood. Familiarity with the basic ideas of time series—autocorrelation, stationarity, and ARMA models—will make the material far more accessible, though it is not strictly required to understand the core logic. No specialist knowledge of finance is needed, although examples are drawn from financial markets. The framework is best learned through application: fitting a model to a financial return series and examining the resulting volatility forecast is the most direct way to appreciate its power.The 8 concepts its tower teaches
- Conditional Heteroskedasticity
- Volatility Clustering
- ARCH Model
- GARCH Model
- EGARCH Model
- GJR-GARCH Model
- IGARCH Model
- Volatility Forecasting
- 1987-Present
Cointegration
About Cointegration
Cointegration is a statistical property of collections of time series variables that describes a long-run equilibrium relationship among them, even when each individual series is non-stationary. The core insight is that while two or more economic variables—such as consumption and income, or prices in different markets—may each wander unpredictably over time, a specific linear combination of them can be stationary. When that happens, the variables are said to be cointegrated: they can drift apart in the short run, but economic forces pull them back toward a stable equilibrium over the longer horizon.The problem that cointegration addresses is the danger of spurious regression. Before the 1980s, economists routinely ran ordinary least squares (OLS) regressions on non-stationary time series data. As Udny Yule had shown as early as 1926, and as Clive Granger and Paul Newbold demonstrated forcefully in their 1974 work, two independent random-walk series can produce a highly significant-looking regression relationship with a high R-squared, even though no genuine causal link exists. Standard detrending techniques do not solve the problem, because many economic series are not trend-stationary but difference-stationary: they contain stochastic trends, meaning a shock to the level is permanent. Charles Nelson and Charles Plosser provided influential statistical evidence in 1982 that many US macroeconomic series—GNP, wages, employment—indeed have such stochastic trends. The implication was stark: the conventional toolkit of the time was systematically producing misleading inferences.The formal framework that resolved this crisis was developed by Robert Engle and Clive Granger in their seminal 1987 paper, which coined the term "cointegration" and introduced the cointegrating vector approach. Their key theoretical contribution was the Granger representation theorem, which established a deep link between cointegration and error correction. If a set of variables is cointegrated, then their dynamic behavior can be equivalently described by an error correction model (ECM). In an ECM, the short-run changes in the variables are regressed on lagged changes plus an error correction term—the lagged deviation from the long-run equilibrium relationship. This term captures the speed at which the system adjusts back toward equilibrium after a shock. The theorem guarantees that cointegration and error correction are two sides of the same coin: cointegration describes the long-run equilibrium, while the error correction term describes the short-run adjustment dynamics that keep the system bound to that equilibrium.To understand the framework, it helps to proceed in a sensible learning order. First, master the distinction between stationary and non-stationary series, particularly the concept of an integrated process of order d, denoted . A series is if it requires one differencing to become stationary. Second, understand why regressions on variables can be spurious. Third, learn the formal definition of cointegration: if X, Y, and Z are each , and there exist coefficients a, b, c such that aX + bY + cZ is integrated of order less than d, then the variables are cointegrated, and the vector (a, b, c) is a cointegrating vector. Fourth, study the Engle-Granger two-step procedure: estimate the long-run equilibrium relationship by OLS, test the residuals for stationarity (typically using an augmented Dickey-Fuller test), and if the residuals are stationary, estimate the error correction model using the lagged residuals as the error correction term. Finally, for systems with more than two variables, the Johansen procedure (developed by Søren Johansen in 1988 and 1991) provides a more general maximum-likelihood framework for estimating and testing multiple cointegrating relationships.The intellectual history of cointegration is tightly bound to the broader shift in time series econometrics during the 1980s. The immediate predecessor framework, Box-Jenkins ARIMA modeling, had provided powerful tools for modeling univariate non-stationary series by differencing, but it treated each series in isolation. Cointegration extended this logic to multivariate systems, asking not just how a single series becomes stationary, but how several non-stationary series can be stationary in combination. This placed the framework in direct dialogue with vector autoregressions (VARs), which model the joint dynamics of multiple series. A VAR in differences loses the long-run information contained in the levels of the variables; a VAR in levels risks spurious results. Cointegration offered the middle path: a vector error correction model (VECM) that includes both the short-run dynamics of a VAR and the long-run equilibrium constraint imposed by the cointegrating relationships. This made cointegration a natural bridge between the univariate ARIMA tradition and the later structural VAR literature, which focused on identifying causal shocks within multivariate systems.The practical prerequisites for understanding cointegration are modest but real. The reader should be comfortable with basic regression analysis, the concept of stationarity, and the idea of an autoregressive process. Familiarity with unit root tests, such as the Dickey-Fuller test, is helpful but not strictly necessary, since the logic of cointegration can be grasped without the formal test statistics. No specialist mathematical training beyond introductory econometrics is required; the framework is fundamentally about interpreting the behavior of linear combinations of trending variables.Cointegration remains central to empirical macroeconomics and finance. It is the standard tool for testing theories that imply long-run relationships, such as purchasing power parity, the Fisher effect, or the link between consumption and income. It is also widely used in financial econometrics for pairs trading and for modeling the joint behavior of asset prices. The framework's enduring significance lies in its resolution of a fundamental dilemma: how to model the short-run dynamics of economic variables without discarding the long-run information that economic theory says should matter. By making the error correction term an explicit part of the model, cointegration gives empirical content to the idea that markets and economies are self-correcting over time, even when individual series appear to wander without bound.The 8 concepts its tower teaches
- Integrated Process
- Spurious Regression
- Cointegration
- Granger Representation Theorem
- Error Correction Model
- Engle-Granger Two-Step Procedure
- Johansen Procedure
- Vector Error Correction Model
- 1990-Present
- 2005-Present
- 2010-Present
How the ideas connect
Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.
Unit Root and Spurious Regression Analysis influenced by Stationary process
The concept of nonstationarity, particularly the random walk, was the direct subject of the spurious regression problem, leading to the formal definition of the unit root.
Carried by Clive Granger; Paul Newbold.
Cointegration grew out of Unit Root and Spurious Regression Analysis
The unit root problem established the necessity of testing for long-run equilibrium, which cointegration formalized by showing that linear combinations of nonstationary series can be stationary.
Carried by Robert Engle; Clive Granger.
Box–Jenkins method influenced by Stationary process
The concept of stationarity, central to the statistical tradition, provided the necessary theoretical prerequisite for the systematic identification of model orders in the Box-Jenkins method.
Carried by Herman Wold.
Bayesian Time Series Analysis influenced by State-Space Models and the Kalman Filter
The state-space representation of time series provided a structural framework that Bayesian methods later adopted to define and estimate latent states.
Carried by Rudolf Kalman.
Cointegration influenced by Vector autoregression
The VAR framework provided the multivariate setting where the concept of cointegration could be applied to systems of multiple time series.
Local Projections influenced by Vector autoregression
The VAR framework established the idea of modeling dynamic relationships between multiple time series, which Local Projections generalized by removing strict parameterization.
Beyond the field
Where Time Series Econometrics's ideas came from in other fields, and what they led to there: the strongest 12 of 48 links.
Stationary process grew out of Stochastic process in Probability Theory
The stationary process is a special class of stochastic process, and the theory of stationary stochastic processes was developed as a branch of the general theory by Khinchin and Kolmogorov in the 1930s. The econometric framework inherits the definition and spectral theory from that mathematical development.
Carried by Aleksandr Khinchin; Andrey Kolmogorov.
State-Space Models and the Kalman Filter influenced Structural Time Series Models in Time Series Analysis
Structural time series models are built by casting unobserved trend, seasonal and irregular components into state-space form and estimating them with the Kalman filter. The state-space/Kalman machinery is the computational core of the structural approach.
Carried by Rudolf Kalman; Andrew Harvey.
State-Space Models and the Kalman Filter influenced by Stochastic estimation and Kalman filtering in Control Systems
Stochastic estimation and Kalman filtering in control systems engineering is the engineering formulation from which the econometric state-space framework was taken over.
Carried by Rudolf E. Kalman.
Unit Root and Spurious Regression Analysis influenced Cointegration and Error Correction in Econometrics
Unit root and spurious regression analysis, especially Granger and Newbold's work and Dickey-Fuller tests, set the problem that cointegration solved: nonstationary series whose regressions are spurious unless a stationary combination exists.
Carried by Clive Granger; Paul Newbold.
Unit Root and Spurious Regression Analysis led to Cointegration in Time Series Analysis
Cointegration grew directly out of the unit-root and spurious-regression literature: Engle and Granger showed that regressions on nonstationary series can be meaningful when the series are cointegrated.
Carried by Robert F. Engle; Clive W. J. Granger.
Vector autoregression led to Structural vector autoregression in Macroeconomics
Structural vector autoregression is a direct refinement of the reduced-form VAR, adding identifying restrictions to recover structural shocks; it grew out of Sims's VAR framework.
Carried by Christopher A. Sims.
Vector autoregression influenced Shrinkage Priors and Bayesian VARs in Bayesian Econometrics
Bayesian VAR shrinkage priors are a technique within time series econometrics' treatment of vector autoregressions. The Minnesota prior and subsequent hierarchical priors were developed as tools for this class of models.
Carried by Robert Litterman.
Cointegration influenced by Vector Autoregression in Time Series Analysis
Engle and Granger built cointegration directly on the vector autoregression framework, showing that cointegrated systems have an error-correction representation within VAR models.
Carried by Robert F. Engle; Clive W. J. Granger.
Cointegration influenced by Vector autoregression in Econometrics
Engle and Granger built cointegration directly on the vector autoregression framework, showing that cointegrated systems have an error-correction representation within VAR models.
Carried by Robert F. Engle; Clive W. J. Granger.
Bayesian Time Series Analysis grew out of Bayesian Econometrics in Econometrics
Bayesian time series analysis is a branch of Bayesian econometrics, applying Bayesian inference to dynamic models; the later framework grew directly out of the earlier field.
Carried by Arnold Zellner.
Local Projections reacted against Vector Autoregression in Time Series Analysis
Local projections were proposed as a direct alternative to vector autoregressions for estimating impulse responses, avoiding the VAR's specification of the full dynamic system.
Carried by Oscar Jorda.
Local Projections reacted against Structural vector autoregression in Macroeconomics
Local projections were introduced by Jorda as an alternative to structural VAR impulse responses, arguing that VARs impose restrictive dynamic assumptions; the method defined itself against SVARs.
Carried by Oscar Jorda.

Time Series Econometrics is a planet in Nousfarer
Its city has a tower for each of these 11 frameworks. Land, walk in, and climb one concept at a time.
Land on Time Series EconometricsRelated fields
Fields its ideas are linked to
- Time Series Analysis 22 links
- Macroeconomics 5 links
- Econometrics 4 links
- Bayesian Econometrics 3 links
- Control Theory 2 links
- Control Systems 2 links
- Stochastic Processes 2 links
- Microeconometrics 1 link
- ML Econometrics 1 link
- Panel Data Methods 1 link
- Asset Pricing Theory 1 link
- Austrian Economics 1 link
More in Econometrics
- Bayesian Econometrics
- Econometrics
- Microeconometrics
- ML Econometrics
- Panel Data Methods
- Structural Estimation
See where Time Series Econometrics sits on the Map of Human Knowledge