EViews
EViews is a Windows-based statistical software package designed primarily for econometric analysis, time series modeling, forecasting, and statistical computations.[1] Developed by Quantitative Micro Software (QMS), a subsidiary of S&P Global since 2010, it features an intuitive, object-oriented user interface that facilitates data management, regression analysis, and visualization for users ranging from academics to financial professionals.[2] First released in March 1994 as version 1.0, EViews succeeded the earlier MicroTSP program, which had been introduced in 1981 for personal computers, marking a shift to a more graphical, Windows-native environment.[1] Originally created by QMS to provide accessible tools for economic research and policy analysis, EViews has evolved through multiple versions, with the latest being EViews 14, which includes enhancements for usability, scripting automation, and integration with tools like Microsoft Excel.[3] It is widely adopted globally, serving over 600 central banks and government agencies, more than 1,600 university economics and business departments (including 78% of the top global universities per U.S. News & World Report rankings), and over 50% of Fortune's Top 100 companies across sectors such as energy, technology, and pharmaceuticals.[1] Key capabilities encompass advanced econometric techniques like vector autoregression (VAR), cointegration analysis, and panel data estimation, alongside forecasting models and graphical outputs for simulation and scenario planning.[2] EViews has maintained its position as a leading tool in empirical economic research, with free student versions available to support education.[1]Overview
Definition and Purpose
EViews is a proprietary, Windows-based software package developed for econometric analysis, encompassing time-series modeling, cross-section data analysis, and panel data econometrics.[4] It provides a comprehensive environment for managing and analyzing economic and financial data through an innovative object-oriented interface that integrates statistical, forecasting, and modeling capabilities.[5] The primary purposes of EViews include handling large datasets, conducting statistical tests, estimating econometric models, generating forecasts, and simulating various economic scenarios.[4] These functionalities enable users to perform rigorous analyses on time-dependent data, cross-sectional observations, and longitudinal panels, supporting tasks from basic descriptive statistics to advanced predictive modeling.[2] EViews evolved from the earlier MicroTSP software, originally released in 1981, adapting its foundational econometric tools for modern computing environments.[1] Targeted at a diverse user base, EViews serves academics and university researchers, professional economists, financial analysts, government agencies, and corporations seeking to derive insights from complex datasets.[5] Its key strengths lie in an intuitive interface that blends spreadsheet-like data manipulation with sophisticated analytical tools, allowing users to quickly import data, visualize results, and automate workflows via programming commands.[4] This design minimizes the learning curve while accommodating both novice and expert users in econometric applications.[6] EViews requires a 64-bit Windows operating system, version 8 or later; recent iterations, such as EViews 14, support Windows 8, 8.1, 10, and 11 (along with certain Windows Server editions).[7] A Pentium or better CPU and at least 512 MB of RAM are recommended, along with sufficient disk space for data storage and installation.[7]Development and Ownership
EViews was originally developed by Quantitative Micro Software (QMS), a company founded in 1984 that specialized in econometric and forecasting software for personal computers.[8] QMS's earlier product, MicroTSP, was one of the first such packages for personal computers, and EViews succeeded it in 1994 as a Windows-based alternative.[9] The ownership of EViews has undergone several transitions reflecting broader consolidations in the data and analytics industry. QMS was acquired by IHS Inc. in May 2010 for $40 million, integrating EViews into IHS's portfolio of information services.[10] In 2016, IHS merged with Markit Ltd. in an all-share transaction valued at over $13 billion to form IHS Markit, enhancing EViews' position within a global provider of financial and economic data.[11] IHS Markit was subsequently acquired by S&P Global in an all-stock deal completed on February 28, 2022, valuing the transaction at approximately $44 billion and positioning EViews under S&P Global's Market Intelligence division.[12] As proprietary software, EViews is distributed under a commercial licensing model that includes single-user licenses for individual professionals, multi-user volume licenses for organizations, and discounted academic editions for educational institutions.[13] A free Student Version Lite is available for learning purposes but imposes restrictions, such as limiting workfiles to three pages, models to ten equations, and total observations to 15,000 across series.[14] EViews receives ongoing support through annual patches and major version updates released by S&P Global, ensuring compatibility with evolving hardware and software environments.[15] Comprehensive documentation, including the PDF guide EViews Illustrated, provides step-by-step tutorials and examples for users.[16] The software also features built-in integrations with third-party data providers, enabling direct access to sources like Thomson Reuters Datastream for financial time series, Haver Analytics for macroeconomic datasets, and CEIC for global economic indicators.[6][17]History
Origins and Early Development
EViews traces its roots to the Time Series Processor (TSP), a pioneering econometric programming language originally developed by Robert Hall during his graduate studies at the Massachusetts Institute of Technology (MIT) in the mid-1960s. Hall initiated work on TSP in 1966 using an IBM 1620 computer, with early development continuing at the University of California, Berkeley, on systems like the CDC 6400. Designed for econometric estimation tasks such as ordinary least squares (OLS) and generalized method of moments (GMM), TSP emphasized numerical accuracy and flexibility, supporting a wide range of models including ARIMA and vector autoregressions (VARs). By the late 1960s and 1970s, TSP had become a staple in academic and research environments, circulating among institutions like Brookings and Wharton, though it was primarily command-line based and suited for mainframe computers.[18] Building on TSP's foundation, MicroTSP emerged as its microcomputer adaptation, first released in 1981 by Quantitative Micro Software (QMS), a company founded by David Lilien. MicroTSP, initially developed for the Apple II and later ported to PCs using compiled BASIC by 1984, introduced graphical and interactive elements to econometric analysis, making it one of the earliest forecasting and analytical packages for personal computers. It addressed the need for accessible time-series tools in the DOS era but remained limited by command-line interfaces and the constraints of early PC hardware. QMS, established in 1980, positioned MicroTSP as a bridge from mainframe TSP to desktop computing, incorporating algorithms influenced by Cambridge University's Department of Applied Economics, including contributions from Lucy Slater and M. Hashem Pesaran.[1][18] In response to the rise of Microsoft Windows and the shortcomings of DOS-based software like MicroTSP—particularly in multitasking, visual data handling, and user-friendly interfaces—QMS launched EViews in January 1994 as version 1.0. This initial release replaced MicroTSP, modernizing the TSP language for Windows environments by introducing a graphical user interface (GUI) that streamlined econometric workflows previously reliant on command-line operations. Early motivations centered on enhancing productivity for economists and analysts through intuitive point-and-click functionality, while retaining compatibility with TSP's core procedures. Version 1.0 focused on foundational capabilities, including basic time-series analysis, regression modeling, and data import from common spreadsheet formats like Lotus 1-2-3 and Excel, thereby democratizing advanced econometrics for academic, corporate, and government users.[1][18]Key Milestones and Evolution
In the mid-1990s, EViews underwent significant expansions through versions 2 to 4, introducing ARIMA modeling capabilities and integration with Microsoft Excel for enhanced data import and export.[19] These additions broadened the software's applicability beyond basic time-series analysis, enabling users to handle cross-sectional and longitudinal datasets more effectively while streamlining workflows with office productivity tools.[20] The 2000s marked further advancements in econometric sophistication, with version 5 released in 2004 introducing state-space models that allowed for flexible representation of dynamic systems with unobserved components, as well as support for panel data analysis. Version 6 in 2007 enhanced forecasting tools, particularly through expanded support for vector autoregression (VAR) models and multivariate GARCH estimation, improving the handling of volatility clustering and multivariate time-series dependencies. These developments solidified EViews as a robust platform for advanced macroeconomic modeling and risk assessment. The 2010s saw pivotal growth following S&P Global's (formerly IHS) acquisition of Quantitative Micro Software in 2010, which facilitated deeper integrations with cloud-based economic databases and third-party data providers like S&P Global Economics (formerly IHS Economics).[10][21] This shift enabled seamless access to global datasets, enhancing real-time analysis for users in government and finance. Version 10, launched in 2017, introduced direct bridging with R and improved Python compatibility, allowing hybrid workflows that combined EViews' econometric strengths with open-source statistical libraries. Entering the 2020s, version 13 in 2022 advanced machine learning tools, including enhanced Bayesian estimation and support for complex non-linear models like non-linear ARDL, alongside better big data handling through expanded data connectivity and seasonal adjustment for high-frequency series.[22] Version 14, released on June 25, 2024, further expanded capabilities with integration of JDemetra+ for seasonal adjustment, Facebook Prophet for forecasting, quantile ARDL estimation, and MIDAS GARCH estimation.[3] These updates emphasized cross-platform compatibility, such as Jupyter Notebook integration, reflecting EViews' evolution from a time-series-centric tool to a comprehensive suite incorporating AI-assisted modeling for diverse analytical needs.[23]User Interface and Accessibility
Graphical User Interface
EViews employs an object-oriented graphical user interface (GUI) that organizes data, models, graphs, and tables into interactive objects, each with dedicated windows featuring multiple views such as spreadsheets, plots, and statistical outputs.[24] This design enables dynamic updating, where changes to underlying data automatically propagate to linked objects like series and models, facilitating efficient workflow management.[24] Spreadsheet-style workfiles serve as the core for viewing and editing data, supporting multi-page structures for handling complex datasets in a familiar, Excel-like format.[6] Navigation within the GUI is primarily menu-driven, offering quick access to analytical tools through context-specific menus—for instance, series objects include options for generating histograms or scatter plots directly—while drag-and-drop capabilities simplify tasks like importing data from external files or rearranging equations.[4] Accessibility is enhanced by keyboard shortcuts for common operations, tooltips that provide contextual help on hover (such as observation details in graphs), and wizard-guided procedures that step users through complex processes, including unit root testing.[6] Multiple windows allow simultaneous viewing and side-by-side comparisons of objects, such as plotting series alongside equation estimates, promoting intuitive exploration.[24] The GUI's visualization tools include built-in graphing for line plots, scatter diagrams, and histograms, with extensive customization options for themes, axes scaling, legends, and backgrounds via templates.[6] Graphs support interactive elements like hovering for data points and can be exported in formats including PNG, JPG, PDF, and Windows metafiles for reporting purposes.[6] Usability in recent versions, such as EViews 14, has been improved with dark mode support to reduce eye strain, enhanced responsiveness for high-resolution and high-DPI displays, and updated dialogs for smoother interactions.[3] For advanced users, the GUI seamlessly integrates with programming features to extend point-and-click operations into scripted workflows.[4]Programming and Command Features
EViews employs a proprietary programming language with syntax loosely modeled after BASIC, enabling users to automate complex analyses through constructs such as loops, conditional statements, and procedures. For instance, loops can iterate over data elements using commands likefor !i = 1 to @rows(vec1), while conditionals support decision-making with if result <> value then followed by endif. Procedures are defined via subroutines, as in subroutine loglike paired with endsub, allowing modular code organization and calls with call subroutine_name. This language also handles matrix operations (e.g., @inverse(mat1) for inversion) and string processing (e.g., @upper(string_var) for case conversion), facilitating data manipulation and custom function creation within an object-oriented framework where objects like series or equations are accessed via _this.[25]
The command window serves as the core interface for both interactive input and batch processing, allowing users to execute sequences of commands for repetitive tasks and ensuring reproducibility through history logging. Commands entered are retained in a session history accessible via CTRL+J (up to the last 30), with full logging enabled by logmode on or output redirection using output(file="log.txt") and spool. Batch execution occurs via program files saved as .prg extensions, invoked with run prog_name.prg or exec prog_name.prg, which supports debugging features like breakpoints and watch windows for step-through execution. This setup is particularly useful for scripting workflows that mirror GUI actions, such as model estimation or data transformations, without manual intervention.[25]
Add-ins and extensions enhance EViews' functionality through user-developed .prg files, registered as reusable tools with the addin command, enabling community-contributed packages for specialized analyses. Integration with external languages supports hybrid workflows: Python sessions open via xopen(type=p), R with xopen(type=r), and MATLAB via xopen(type=m), allowing data transfer with xput and xget or direct command execution like xrun("python_script.py"). These interfaces leverage COM automation for bidirectional communication, such as passing EViews workfiles to R for advanced plotting or importing MATLAB matrices into EViews series. Program objects further automate tasks by encapsulating custom functions (e.g., program custom_est for tailored estimations) and an API exposes EViews objects to external applications via methods like fetch or solve for model solving.[6][25]
Despite these capabilities, EViews' programming environment is object-oriented but specialized for econometric and statistical operations, lacking the breadth of general-purpose languages like Python for non-data tasks such as web scraping or low-level system programming.[25]
Data Management
Supported Formats and Compatibility
EViews utilizes a proprietary workfile format with the .wf1 extension to store structured data, including time series, cross-sections, groups of series, models, and associated metadata, ensuring efficient handling of econometric datasets.[26] This format has remained compatible across versions since EViews 1, allowing seamless backward and forward compatibility for workfiles.[27] For import compatibility, EViews natively supports a wide array of formats from popular statistical and spreadsheet software, including Microsoft Excel files (.xlsx and .xlsm), comma-separated values (CSV) as formatted ASCII text, SPSS native (.sav) and portable files, SAS transport files (.xpt), Stata files (.dta), Gauss dataset files, RATS and TSP formats, and binary files.[19][28] Additionally, EViews provides direct connectivity to online economic databases such as the International Monetary Fund (IMF), World Bank, Federal Reserve Economic Data (FRED), Organisation for Economic Co-operation and Development (OECD), Bloomberg, and DBNomics (as of 2025), allowing users to fetch and import data with a few clicks via built-in interfaces, including an enhanced graphical user interface for SDMX databases introduced in EViews 14.[29][30] It also offers ODBC and JDBC connectivity for direct access to relational databases such as Microsoft Access and SQL Server, though full ODBC support is limited to the Enterprise Edition.[19] Other formats like HTML tables and Tableau files are also importable, facilitating integration with diverse data sources.[19] Export options mirror most import formats, enabling output to Excel, CSV, SPSS, SAS, Stata, Gauss, RATS, TSP, and binary files for interoperability with other analysis tools.[28] EViews further supports exporting reports and outputs in PDF and HTML formats to preserve formatting for documentation and sharing.[28] Partial compatibility exists with open-source software like Gretl, which can read certain EViews workfiles, though full feature preservation may require conversion.[19] EViews enhances cross-software interoperability through built-in bridges to R and Python, allowing users to launch external sessions, transfer data bidirectionally, and execute commands between environments for advanced scripting and analysis.[31] Third-party packages such as pyeviews and EviewsR further streamline data exchange by enabling Python and R scripts to interact directly with EViews workfiles and objects.[32][33] A notable limitation is EViews' lack of native support for macOS or Linux operating systems in full commercial versions (Standard and Enterprise), restricting them to Windows; however, University and Student editions offer native macOS support (compatible with macOS Catalina and later, excluding macOS 15 Sequoia as of 2024), and users can run full versions on non-Windows platforms via virtual machines or emulators, though performance may vary.[14][34]Import, Export, and Manipulation Tools
EViews provides a suite of tools for importing data into workfiles, supporting both interactive wizards and command-line procedures to handle structured and unstructured sources. The import wizard, accessible via File > Open > Foreign Data as Workfile, guides users through selecting data ranges, specifying variable names, and configuring options for various inputs, such as spreadsheets or text files.[35] For unstructured data, drag-and-drop functionality allows direct import from compatible files, with the software automatically detecting and structuring observations based on dates or indices.[35] Frequency conversion is integrated into the import process, enabling seamless aggregation or disaggregation, such as converting daily observations to monthly averages using methods like sum, average, first/last, or interpolation techniques (e.g., linear or spline).[35] Thewfopen command automates this, as in wfopen type=excel file="data.xlsx" range "Sheet1!A1:B10", which opens and structures the data into a dated workfile.[25]
Export workflows in EViews facilitate sharing results and data through batch operations and automation. Users can export workfiles, series groups, tables, and graphs via File > Save As, selecting formats like Excel or CSV, with options for including metadata or ranges.[35] Batch export is supported for multiple objects, such as saving spool outputs (e.g., regression tables or plots) to disk in RTF, PDF, or CSV using Proc > Save Spool to Disk, ideal for reports.[35] Automation occurs through programs (.prg files) executed via the command window, employing wfsave for scripted exports, as in wfsave excel output.xlsx to generate spreadsheets dynamically.[25] This enables workflows like exporting updated forecasts in loops, integrating with external tools via OLEDB drivers.[35]
Data manipulation tools in EViews center on series transformations and handling for econometric preparation. Basic operations include logarithmic transformations (log(series) or @log), lags (series(-n) or @lag(series, n)), and differences (d(series) for first-order or dlog for log differences), applied interactively via Quick > Show or via commands like series lgdp = [log](/page/Log)(gdp).[25][35] Pooling for panel data involves creating pool objects from multi-page workfiles, specifying cross-sections and time dimensions (e.g., pool mypool followed by estimation), which aggregates unbalanced panels while preserving identifiers.[35] Missing value handling uses conditional statements and functions like @isna(series) to detect NAs, with interpolation options (e.g., linear via series interp = series if !@isna(series) else @trend(series, 1) or built-in procs like cubic spline), and sample exclusion via smpl if !@isna(series).[35][25]
Workfiles serve as the core data structure in EViews, organizing series with dated or frequency-indexed observations (e.g., annual, quarterly, monthly, or irregular). Created via File > New > Workfile or wfcreate page1 q 2000q1 2020q4, they support multi-page setups for panels, where each page holds cross-sectional units.[25][35] Groups organize subsets of series for efficient manipulation, formed interactively (Object > New Object > Group) or via group g1 x y z, allowing operations like averaging (@rmean) across members without altering originals.[25][35]
Advanced features enhance subsetting and integration of multi-source data. The smpl command defines observation subsets for all subsequent operations, such as smpl 1955m1 1958m12 if condition, restricting analyses to specific ranges or criteria while preserving the full workfile.[25] Merge and join operations use fetch to pull series from databases (fetch(d=mydata) gdp) or linkto for dynamic connections (linkto dbname::series), enabling appends across workfiles via Proc > Append or append mydata.wf1 for combining panels.[25][35] These tools support key matching on identifiers, facilitating joins from disparate sources like economic databases.[35]
Statistical Analysis
Descriptive Statistics and Diagnostics
EViews provides a suite of tools for computing descriptive statistics and performing diagnostic tests on series, groups, and workfiles, facilitating initial data exploration prior to more advanced analysis. Users can access these features through intuitive quick menus under the View menu, such as View > Descriptive Statistics & Tests, or via command-line instructions like thestats command, which generates comprehensive summaries including means, medians, standard deviations, variances, minima, maxima, skewness, and kurtosis for individual series or groups of variables.[35] For correlations, EViews supports multiple measures including Pearson, Spearman, and Kendall's tau coefficients, computed via View > Descriptive Statistics & Tests > Correlation or the cor command, allowing assessment of linear and rank-based relationships among variables.[35] Graphical representations enhance these summaries; histograms, accessible through View > Descriptive Statistics & Tests > Histogram and Stats or the graph command, display frequency distributions with overlaid density estimates, while box plots, available under View > Graph > Boxplot, visualize medians, quartiles, and potential outliers using interquartile range (IQR) thresholds of 1.5 or 3.0. EViews 14 enhances these with options like average shifted histograms and kernel density improvements.[35][6][36]
Diagnostic tests in EViews focus on key assumptions for econometric analysis, integrated into the descriptive framework to identify data issues early. The Jarque-Bera test for normality, performed via View > Descriptive Statistics & Tests > Normality Test or within histogram outputs, evaluates skewness and kurtosis against a chi-squared distribution, providing a p-value to assess deviation from normality; for example, on a series like GDP growth, it might yield a statistic of 5.23 with p=0.073, indicating marginal non-normality.[35] Heteroskedasticity can be checked using the White test through View > Descriptive Statistics & Tests > Heteroskedasticity Tests > White, which regresses squared residuals on cross-products of regressors to test constant variance, outputting an F-statistic and p-value for decision-making.[35] Multicollinearity is diagnosed via Variance Inflation Factors (VIF), computed under View > Descriptive Statistics & Tests > Covariance Analysis or as an option in correlation matrices, where VIF values exceeding 10 signal high collinearity among explanatory variables; EViews displays these in tables alongside tolerance values (1/VIF).[35] Automatic outlier detection integrates with these tools, flagging extremes in box plots or via value-based coloring in spreadsheet views (View > Spreadsheet Options), and descriptive tables routinely include p-values for tests to aid interpretation. EViews 14 adds series-based outlier detection using methods like Tukey fences, wavelet, and ARMA.[37][36]
For panel and cross-section data, EViews offers specialized group statistics and balance checks to handle structured datasets. Group statistics, accessed via View > Group Statistics or the groupstats command, compute summaries by classification variables, such as means and variances across cross-sectional units or time periods, with options for common or individual samples.[35] Balance checks, under Proc > Structure or workfile settings, verify panel completeness by identifying missing observations and allowing creation of balanced subsets via Proc > Make Balanced Panel, ensuring equitable representation across groups.[35] These features support exploratory analysis in multi-dimensional data without delving into time-series specifics like stationarity.[36]
Output from descriptive statistics and diagnostics is presented in customizable tables, which users can freeze (via the Freeze button) or edit directly for tailored reporting, including p-values and confidence intervals.[35] Exports are straightforward, with options to save tables and graphs to Word or Excel formats through File > Export > To RTF/Excel or right-click Save to Disk, enabling seamless integration into reports or further processing in other software.[35] This flexibility ensures that exploratory insights are readily documented and shared.[36]
Stationarity Testing
Stationarity refers to a property of time series data where the statistical characteristics, such as mean and variance, remain constant over time, which is crucial for valid econometric inference to avoid spurious regressions.[38] EViews provides a range of unit root tests to assess stationarity, including the Augmented Dickey-Fuller (ADF) test, which augments the basic Dickey-Fuller regression with lagged difference terms to account for higher-order autoregressive processes under the null hypothesis of a unit root.[38] The Phillips-Perron (PP) test modifies the ADF by using a non-parametric correction for serial correlation and heteroskedasticity, making it robust to unspecified error structures while maintaining the same asymptotic distribution as the Dickey-Fuller test.[39][38] In contrast, the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test reverses the hypothesis, testing the null of stationarity against the alternative of a unit root or trend-stationarity, providing a complementary perspective to ADF and PP results.[38] The Elliott-Rothenberg-Stock (ERS) point-optimal test, including its DF-GLS variant, offers improved power by de-trending the series via generalized least squares before applying a point-optimal statistic targeted at local alternatives to the unit root. EViews 14 adds the Ng-Perron test for enhanced unit root assessment.[40][38][6] Implementation in EViews allows users to select these tests via the series' View > Unit Root Test menu, with automatic lag selection based on information criteria such as AIC, SIC, or Hannan-Quinn, or modified versions thereof, to balance model fit and parsimony; the maximum lag is often set to a function of sample size, like \min(T/3, 12) \times (T/100)^{0.25}.[38] Options for deterministic components include no intercept, constant only, linear trend, or quadratic trend, enabling tests for different forms of non-stationarity such as random walks with or without drift. EViews 14 also supports unit root tests with breakpoints and seasonal unit root tests.[38][6] Graphical outputs, such as residual plots and test statistic distributions, facilitate visual inspection of results, particularly for cointegration views in multivariate contexts.[38] To address non-stationarity, EViews supports differencing transformations using thed(series, order) command for first- or second-order differencing to induce stationarity in integrated series, and seasonal differencing via d(series, 0, period) for adjustments like quarterly data with period 4.[38] Additional tools include built-in seasonal adjustment procedures, such as X-13 ARIMA-SEATS, to remove seasonal patterns while preserving underlying trends.[38] EViews 14 introduces trend testing with parametric and non-parametric methods, including bootstrapping options.[37]
For multivariate time series, EViews implements the Johansen cointegration test to detect long-run equilibrium relationships among non-stationary variables, using trace and maximum eigenvalue statistics to determine the cointegrating rank within a vector autoregression (VAR) framework, with options for linear trends or constants in the cointegrating relations. EViews 14 improves cointegration testing with added stability diagnostics like the Hansen test.[38][6] The test is accessed through Quick > Estimate VAR or View > Cointegration Test on a group object, supporting restrictions on cointegrating vectors and providing graphical views of eigenvalues and adjustment coefficients for interpretation.[38]
Econometric Modeling
Regression and Estimation Methods
EViews provides a suite of tools for estimating single-equation regression models, emphasizing ease of specification and robust inference for econometric analysis. Core to these capabilities are equation objects, which encapsulate model estimation, diagnostics, and forecasting in a unified interface. Users can specify models via dialog boxes or command lines, supporting both linear and nonlinear functional forms while integrating options for covariance adjustments to handle common violations like heteroskedasticity and autocorrelation.[41][6] Ordinary least squares (OLS) serves as the foundational method for linear regression in EViews, estimating coefficients by minimizing the sum of squared residuals according to the model y = X\beta + \epsilon, where \hat{\beta} = (X'X)^{-1}X'y. Equation objects facilitate OLS estimation through simple specifications likels y c x1 x2, producing outputs including coefficient estimates, standard errors, t-statistics, R-squared, and F-statistics. To address heteroskedasticity, EViews offers robust (Huber-White) standard errors, computed via sandwich estimators that adjust the covariance matrix without altering point estimates. For time-series data prone to autocorrelation, heteroskedasticity and autocorrelation consistent (HAC) covariance matrices, such as Newey-West estimators, provide consistent inference by prewhitening residuals and selecting bandwidths automatically or via user input. These features ensure reliable hypothesis testing even under assumption violations, as demonstrated in consumption function models where adjusted R-squared values often exceed 0.9.[6]
For nonlinear models, EViews employs nonlinear least squares (NLS) estimation, which iteratively minimizes the sum of squared residuals for implicit functions like CES production specifications. Algorithms such as Gauss-Newton, BFGS, or Marquardt are available, with starting values set via parameter objects to aid convergence. NLS supports ARMA error structures and threshold models, making it suitable for implicit relationships where linearity fails. Binary outcome models, common in econometrics for dichotomous dependent variables, are estimated using maximum likelihood methods like logit (logistic distribution) or probit (normal distribution). These produce odds ratios, predicted probabilities, and pseudo-R-squared measures; for instance, logit specifications yield log-likelihood values for model comparison, with options for clustered standard errors. EViews 14 introduces additional regularization techniques, including elastic net, ridge regression, and LASSO, for handling multicollinearity and variable selection in high-dimensional regressions.[6]
Panel data estimation in EViews extends OLS to cross-sectional time-series structures, offering pooled OLS for unrestricted averaging across units, fixed effects to control for unobserved individual heterogeneity via within-group transformations, and random effects using generalized least squares (GLS) under the assumption that effects are uncorrelated with regressors. The Hausman test, integrated post-estimation, helps select between fixed and random effects, with examples showing chi-square statistics around 800 for rejection of random effects. Pooled OLS ignores panel structure but serves as a baseline, while fixed effects often yield within-R-squared values above 0.7 in hedonic pricing applications. Version 14 adds support for Difference-in-Difference estimation in panel contexts.[6]
Post-estimation diagnostics are seamlessly integrated into equation objects, allowing immediate assessment of model adequacy. The Breusch-Pagan test detects heteroskedasticity by regressing squared residuals on explanatory variables, reporting LM chi-square statistics (e.g., exceeding 1000 in large samples indicates violation). For serial correlation, the Durbin-Watson statistic evaluates first-order autocorrelation, with values near 2 suggesting no issue; higher-order extensions are available via Breusch-Godfrey tests. These tools, alongside variance inflation factors for multicollinearity, enable iterative refinement without leaving the equation interface. Prior to time-series regression, users may briefly reference stationarity tests to ensure cointegration assumptions hold. EViews 14 enhances estimation efficiency with updated algorithms and greater control over coefficient penalties.[41][6]
Equation specification in EViews supports flexible model building, including stepwise variable selection for automated inclusion/exclusion based on p-value thresholds (forward, backward, or bidirectional). Restricted models impose linear constraints on coefficients, such as equality or zero restrictions, tested via Wald statistics with chi-square p-values for joint significance. These options, applied via the estimation dialog or commands like @restrict, facilitate hypothesis-driven analysis while maintaining computational efficiency.[6]
Time Series and Panel Data Models
EViews provides robust tools for estimating time series models that capture temporal dependencies in data, building on foundational autoregressive integrated moving average (ARIMA) frameworks originally developed by Box and Jenkins.[42] These models, including ARMA variants, allow users to specify autoregressive (AR) and moving average (MA) orders, such as AR(p) and MA(q), with options for differencing to achieve stationarity and seasonal components via sar and sma terms.[6] In EViews, ARIMA estimation employs maximum likelihood (ML), conditional least squares (CLS), or unconditional least squares methods, supporting dynamic lag structures like ar(1 to 5) and handling missing values through sample adjustment.[6] Vector autoregression (VAR) and vector error correction models (VECM) extend univariate approaches to multivariate settings, as pioneered by Sims for VAR analysis of macroeconomic interdependencies.[43] EViews facilitates VAR specification with flexible lag selections, often determined via information criteria like AIC or SIC, and incorporation of exogenous variables or deterministic terms such as trends.[6] For VECM, which addresses cointegration as formalized by Johansen, users can define cointegrating equations and test for long-run relationships using ML estimation.[44] Seasonal dummies are easily added via the @seas function or custom series, enabling adjustment for periodic patterns in economic data.[6] Impulse response functions (IRFs) in VAR/VECM, computed through orthogonalization or structural identification, illustrate shock propagation across variables, aiding analysis of policy impacts. EViews 14 introduces mixed frequency VARs and Markov Switching VARs for more advanced multivariate time series analysis.[6] Volatility modeling in EViews draws from ARCH processes introduced by Engle and generalized to GARCH by Bollerslev, essential for capturing time-varying variance in financial and economic series.[45][46] The software supports ARCH(p), GARCH(p,q), and extensions like EGARCH, TARCH, and component GARCH, with variance equations incorporating lagged residuals and conditional variances, plus exogenous regressors or ARCH-M terms for risk-return tradeoffs.[6] Estimation relies on ML with distributions such as normal, Student's t, or generalized error, and includes diagnostics like ARCH LM tests for residual heteroskedasticity.[6] These models generate conditional variance forecasts, crucial for risk assessment in volatile markets. For panel data, EViews implements dynamic models using generalized method of moments (GMM), as advanced by Hansen, to handle endogeneity and fixed effects in grouped time series.[47] The Arellano-Bond estimator, a one- or two-step GMM approach for first-differenced dynamic panels, uses lagged levels as instruments for differenced variables, accommodating autoregressive structures via the @dyn keyword.[48][6] Instrumental variables (IV) methods, including two-stage least squares (TSLS), allow specification of instruments with @inst or @stackinst, supporting tests for weak instruments and endogeneity via Hausman statistics.[6] Lag structures in panels mirror time series options, with serial correlation tests like Arellano-Bond ensuring model validity.[6] Estimation across these models emphasizes ML for precise parameter recovery in nonlinear settings, with EViews offering optimization algorithms like BFGS or Newton-Raphson.[6] Bayesian methods are available for Bayesian VAR (BVAR), employing priors such as Minnesota or independent normal-Wishart, and Gibbs sampling for posterior inference in select time series contexts.[6] In applications, these tools forecast economic indicators like GDP or inflation by modeling persistence and cointegration in VAR/VECM or ARIMA setups, while GARCH variants quantify risk through volatility clustering in asset returns or exchange rates.[6] Dynamic panel GMM further enables cross-country growth analysis, addressing unobserved heterogeneity for reliable policy insights. EViews 14 also supports quantile regression and quantile ARDL models for panel and time series data.[6]Forecasting and Simulation
Forecasting Techniques
EViews supports static and dynamic forecasting methods to generate in-sample and out-of-sample predictions from estimated equations and vector autoregression (VAR) models. Static forecasting employs actual observed values of explanatory variables throughout the forecast horizon, making it suitable for evaluating model fit within the sample period or short-term projections where data availability is high. In contrast, dynamic forecasting iterates using previously forecasted values for lagged dependent variables, which is essential for longer horizons or out-of-sample predictions where future explanatory data are unavailable. These approaches allow users to assess model performance by comparing predictions against holdout data, with EViews automatically computing standard errors for forecasts to quantify prediction uncertainty.[49] For short-term forecasts, EViews incorporates exponential smoothing techniques, such as simple, Holt's linear trend, and Holt-Winters seasonal methods, which apply weighted averages to past observations to produce smoothed predictions. These are particularly effective for univariate time series with trends or seasonality, enabling quick implementation via the software's Proc/Smoothing procedure. Additionally, autoregressive conditional heteroskedasticity (ARCH) and generalized ARCH (GARCH) models facilitate volatility forecasting in financial and economic series, where conditional variances are projected alongside mean forecasts to capture time-varying uncertainty. Users can estimate these models and generate forecasts directly from equation objects, often yielding reliable short-term predictions for assets with clustered volatility.[49][50] Scenario-based forecasting in EViews enables users to explore alternative predictions by imposing different assumptions on exogenous variables, such as policy shocks or economic perturbations, within estimated models. This involves specifying bind conditions or additive adjustments to variables, allowing for what-if analyses that generate tailored forecast paths without re-estimating the underlying model. To evaluate forecast quality, EViews computes accuracy measures including root mean squared error (RMSE), mean absolute error (MAE), and Theil's U inequality coefficient, which decomposes error into bias, variance, and covariance proportions to diagnose sources of inaccuracy. Fan charts further visualize uncertainty by plotting forecasts with confidence bands derived from historical errors or model variances, providing a probabilistic view of future outcomes.[19][49] Integration of forecasting occurs through model objects in EViews, which link estimation results from multiple equations into a cohesive framework for generating unified predictions. These objects facilitate dynamic updates between estimation and forecast views, supporting scenario overrides and automatic propagation of changes across linked components. This setup streamlines the transition from time-series model fitting to predictive applications, ensuring consistency in multi-equation environments like macroeconomic simulations.[49]Model Simulation and Scenario Analysis
EViews provides robust tools for model simulation, enabling users to generate projected paths for endogenous variables based on estimated econometric models. The software's Model object facilitates the creation of systems comprising equations, identities, and exogenous variables, which are solved using iterative algorithms such as Gauss-Seidel, Newton, or Broyden's method to achieve convergence in dynamic or static simulations.[51] These simulations build on forecasting baselines by incorporating variability to assess uncertainty and policy impacts.[38] Stochastic simulation in EViews employs Monte Carlo methods to introduce random shocks to model residuals, allowing for the generation of error bands and probability distributions around forecast paths. Users specify the number of repetitions—typically hundreds or thousands—to approximate the distribution of outcomes, which helps quantify statistical and formulation errors in multivariate systems.[51] For instance, in a time series model, stochastic draws from estimated residual variances produce fan charts illustrating confidence intervals for variables like GDP growth under uncertain conditions.[52] Scenario analysis tools enable the exploration of hypothetical "what-if" situations by binding or exogenizing variables, such as fixing interest rates to evaluate monetary policy effects. Through the View/Scenarios dialog or commands likecontrol and fliptype, users override endogenous variables (e.g., setting an exogenous shock to money supply) or add alternative paths for exogenous inputs, comparing outcomes across multiple scenarios labeled with suffixes like _1 or _2.[38] This is particularly useful for policy simulations, where a baseline scenario might assume steady exogenous growth, while alternatives test deviations like a 5% increase in fiscal spending.[52]
System simulation extends these capabilities to interconnected equations, capturing economy-wide effects through linked variables in structural models. The solve command processes the entire system, propagating shocks across equations to trace dynamic responses, including multiplier effects in macroeconomic frameworks.[51] Impulse response analysis, integrated within model or VAR objects, computes the time path of variables to a one-time shock (e.g., a unit increase in oil prices), with standard errors derived from analytical methods or Monte Carlo replications for robustness.[53]
Output from simulations is visualized through customizable tables displaying variable paths, means, and percentiles across scenarios or repetitions, alongside graphs such as line plots of medians with error bands. Probability distributions can be output as histograms or cumulative density functions to highlight risks, like the likelihood of negative growth under stochastic draws.[38]
Advanced features include bootstrap resampling for confidence intervals in simulation outputs, where residuals are resampled with replacement to simulate variability in non-standard models, such as functional coefficient regressions. This method enhances reliability by accounting for parameter uncertainty without assuming normality.[38]