Linear models, smooth transition autoregressions, and neural networks for forecasting macroeconomic time series: A re-examination

In this paper, we examine the forecast accuracy of linear autoregressive, smooth transition autoregressive (STAR), and neural network (NN) time series models for 47 monthly macroeconomic variables of the G7 economies. Unlike previous studies that typically consider multiple but fixed model specifications, we use a single but dynamic specification for each model class. The point forecast results indicate that the STAR model generally outperforms linear autoregressive models. It also improves upon several fixed STAR models, demonstrating that careful specification of nonlinear time series models is of crucial importance. The results for neural network models are mixed in the sense that at long forecast horizons, an NN model obtained using Bayesian regularization produces more accurate forecasts than a corresponding model specified using the specific-to-general approach. Reasons for this outcome are discussed.     

Selecting a Nonlinear Time Series Model using Weighted Tests of Equal Forecast Accuracy*

Nonlinear time series models have become fashionable tools to describe and forecast a variety of economic time series. A closer look at reported empirical studies, however, reveals that these models apparently fit well in‐sample, but rarely show a substantial improvement in out‐of‐sample forecasts, at least over linear models. One of the many possible reasons for this finding is the use of inappropriate model selection criteria and forecast evaluation criteria. In this paper we therefore propose a novel criterion, which we believe does more justice to the very nature of nonlinear models. Simulations show that this criterion outperforms those criteria currently in use, in the sense that the true nonlinear model is more often found to perform better in out‐of‐sample forecasting than a benchmark linear model. An empirical illustration for US GDP emphasizes its relevance.     

Forecasting performances of three automated modelling techniques during the economic crisis 2007–2009

In this work we consider the forecasting of macroeconomic variables during an economic crisis. The focus is on a specific class of models, the so-called single hidden-layer feed-forward autoregressive neural network models. What makes these models interesting in the present context is the fact that they form a class of universal approximators and may be expected to work well during exceptional periods such as major economic crises. Neural network models are often difficult to estimate, and we follow the idea of White (2006) of transforming the specification and nonlinear estimation problem into a linear model selection and estimation problem. To this end, we employ three automatic modelling devices. One of them is White’s QuickNet, but we also consider Autometrics, which is well known to time series econometricians, and the Marginal Bridge Estimator, which is better known to statisticians. The performances of these three model selectors are compared by looking at the accuracy of the forecasts of the estimated neural network models. We apply the neural network model and the three modelling techniques to monthly industrial production and unemployment series from the G7 countries and the four Scandinavian ones, and focus on forecasting during the economic crisis 2007–2009. The forecast accuracy is measured using the root mean square forecast error. Hypothesis testing is also used to compare the performances of the different techniques.     

Predicting economic contractions and expansions with the aid of professional forecasts

Traditional econometric models of economic contractions typically perform poorly in forecasting exercises. This criticism is also frequently levelled at professional forecast probabilities of contractions. This paper addresses the problem of incorporating the entire distribution of professional forecasts into an econometric model for forecasting contractions and expansions. A new augmented probit approach is proposed, involving the transformation of the distribution of professional forecasts into a ‘professional forecast’ prior for the economic data underlying the probit model. Since the object of interest is the relationship between the distribution of professional forecasts and the probit model’s economic-data dependent parameters, the solution avoids criticisms levelled at the accuracy of professional forecast based point estimates of contractions. An application to US real GDP data shows that the model yields significant forecast improvements relative to alternative approaches.     

Realized volatility forecasting and market microstructure noise

We extend the analytical results for reduced form realized volatility based forecasting in ABM (2004) to allow for market microstructure frictions in the observed high-frequency returns. Our results build on the eigenfunction representation of the general stochastic volatility class of models developed byMeddahi (2001). In addition to traditional realized volatility measures and the role of the underlying sampling frequencies, we also explore the forecasting performance of several alternative volatility measures designed to mitigate the impact of the microstructure noise. Our analysis is facilitated by a simple unified quadratic form representation for all these estimators. Our results suggest that the detrimental impact of the noise on forecast accuracy can be substantial. Moreover, the linear forecasts based on a simple-to-implement ‘average’ (or ‘subsampled’) estimator obtained by averaging standard sparsely sampled realized volatility measures generally perform on par with the best alternative robust measures.     

Tests for model specification in the presence of alternative hypotheses: Some further results

In Davidson and MacKinnon (1981), two of the present authors proposed a novel and very simple procedure for testing the specification of a nonlinear regression model against the evidence provided by a non-nested alternative. In this paper we extend their results in several directions. First, we relax a number of the assumptions of the previous paper; we admit the possibility that the nonlinear regression functions may depend on lagged dependent variables, and we do not require that the error terms be normally distributed. Second, we show how the earlier procedure may straightforwardly be generalized to the case where the two non-nested models involve different transformations of the dependent variable. Finally, we propose a simple procedure for testing non-nested linear regression models which have endogenous variables on the right-hand side, and have therefore been estimated by two-stage least squares.     

Modelling and forecasting government bond spreads in the euro area: A GVAR model

This paper proposes an extension to Global Vector Autoregressive (GVAR) models to capture time-varying interdependence among financial variables. Government bond spreads in the euro area feature a time-varying pattern of co-movement that poses a serious challenge for econometric modelling and forecasting. This pattern of the data is not captured by the standard specification that model spreads as persistent processes reverting to a time-varying mean determined by two factors: a local factor, driven by fiscal fundamentals and growth, and a global world factor, driven by the market’s appetite for risk. This paper argues that a third factor, expectations of exchange rate devaluation, gained traction during the crises. This factor is well captured via a GVAR that models the interdependence among spreads by making each country’s spread function of global European spreads. Global spreads capture the exposure of each country’s spread to other spreads in the euro area in terms of the time-varying ‘distance’ between their fiscal fundamentals. This new specification dominates the standard one in modelling the time-varying pattern of co-movements among spreads and the response of euro area spreads to the Greek debt crisis.     

Forecasting China's foreign exchange reserves using dynamic model averaging: The roles of macroeconomic fundamentals,financial stress and economic uncertainty

We develop models for examining possible predictors of growth of China's foreign exchange reserves that embrace Chinese and global trade, financial and risk (uncertainty) factors. Specifically, by comparing with other alternative models, we show that the dynamic model averaging (DMA) and dynamic model selection (DMS) models outperform not only linear models (such as random walk, recursive OLS-AR(1) models, recursive OLS with all predictive variables models) but also the Bayesian model averaging (BMA) model for examining possible predictors of growth of those reserves. The DMS is the best overall across all forecast horizons. While some predictors matter more than others over the forecast horizons, there are few that stand the test of time. The US–China interest rate differential has a superior predictive power among the 13 predictors considered, followed by the nominal effective exchange rate and the interest rate spread for most of the forecast horizons. The relative predictive prowess of the oil and copper prices alternates, depending on the commodity cycles. Policy implications are also provided.     

A nonlinear panel data model of cross-sectional dependence

This paper proposes a nonlinear panel data model which can endogenously generate both ‘weak’ and ‘strong’ cross-sectional dependence. The model’s distinguishing characteristic is that a given agent’s behaviour is influenced by an aggregation of the views or actions of those around them. The model allows for considerable flexibility in terms of the genesis of this herding or clustering type behaviour. At an econometric level, the model is shown to nest various extant dynamic panel data models. These include panel AR models, spatial models, which accommodate weak dependence only, and panel models where cross-sectional averages or factors exogenously generate strong, but not weak, cross sectional dependence. An important implication is that the appropriate model for the aggregate series becomes intrinsically nonlinear, due to the clustering behaviour, and thus requires the disaggregates to be simultaneously considered with the aggregate. We provide the associated asymptotic theory for estimation and inference. This is supplemented with Monte Carlo studies and two empirical applications which indicate the utility of our proposed model as a vehicle to model different types of cross-sectional dependence.     

Chi-squared tests for evaluation and comparison of asset pricing models

This paper presents a general statistical framework for estimation, testing and comparison of asset pricing models using the unconstrained distance measure of Hansen and Jagannathan (1997). The limiting results cover both linear and nonlinear models that could be correctly specified or misspecified. We propose modified versions of the existing model selection tests and new pivotal specification and model comparison tests with improved finite-sample properties. In addition, we provide formal tests of multiple model comparison. The excellent size and power properties of the proposed tests are demonstrated using simulated data from linear and nonlinear asset pricing models.     

How much nominal rigidity is there in the US economy? Testing a new Keynesian DSGE model using indirect inference

We evaluate the Smets-Wouters New Keynesian model of the US postwar period, using indirect inference, the bootstrap and a VAR representation of the data. We find that the model is strongly rejected. While an alternative (New Classical) version of the model fares no better, adding limited nominal rigidity to it produces a ‘weighted’ model version closest to the data. But on data from 1984 onwards - the ‘great moderation’ - the best model version is one with a high degree of nominal rigidity, close to New Keynesian. Our results are robust to a variety of methodological and numerical issues.     

Estimating panel data models in the presence of endogeneity and selection

We consider estimation of panel data models with sample selection when the equation of interest contains endogenous explanatory variables as well as unobserved heterogeneity. Assuming that appropriate instruments are available, we propose several tests for selection bias and two estimation procedures that correct for selection in the presence of endogenous regressors. The tests are based on the fixed effects two-stage least squares estimator, thereby permitting arbitrary correlation between unobserved heterogeneity and explanatory variables. The first correction procedure is parametric and is valid under the assumption that the errors in the selection equation are normally distributed. The second procedure estimates the model parameters semiparametrically using series estimators. In the proposed testing and correction procedures, the error terms may be heterogeneously distributed and serially dependent in both selection and primary equations. Because these methods allow for a rather flexible structure of the error variance and do not impose any nonstandard assumptions on the conditional distributions of explanatory variables, they provide a useful alternative to the existing approaches presented in the literature.     

Forecasting with equilibrium-correction models during structural breaks

When location shifts occur, cointegration-based equilibrium-correction models (EqCMs) face forecasting problems. We consider alleviating such forecast failure by updating, intercept corrections, differencing, and estimating the future progress of an ‘internal’ break. Updating leads to a loss of cointegration when an EqCM suffers an equilibrium-mean shift, but helps when collinearities are changed by an ‘external’ break with the EqCM staying constant. Both mechanistic corrections help compared to retaining a pre-break estimated model, but an estimated model of the break process could outperform. We apply the approaches to EqCMs for UK M1, compared with updating a learning function as the break evolves.     

Real-time macroeconomic forecasting with leading indicators: An empirical comparison

This paper demonstrates that the Conference Board’s Composite Leading Index (CLI) has significant real-time predictive ability for Industrial Production (IP) growth rates at horizons from one to six months ahead over the period 1989-2009. A popular but unrealistic analysis, which combines real-time data for CLI and final vintage data for IP as predictor variables, obscures the actual predictive content of the CLI, in the sense that in that case, the improvements in forecast accuracy relative to a univariate AR model are not significant. The CLI appears to be less useful for forecasting growth rates of the Conference Board’s Composite Coincident Index (CCI) in real time, as a univariate AR model performs better. This result is mostly due to its disappointing performance during the first five years of the forecast period. The CLI may not be the best way of exploiting the information contained in the underlying individual leading indicator variables. The use of principal components instead of CLI leads to more accurate real-time forecasts for both IP and CCI growth rates.     

Markov-switching models with endogenous explanatory variables II: A two-step MLE procedure

This paper proposes a two-step maximum likelihood estimation (MLE) procedure to deal with the problem of endogeneity in Markov-switching regression models. A joint estimation procedure provides us with an asymptotically most efficient estimator, but it is not always feasible, due to the ‘curse of dimensionality’ in the matrix of transition probabilities. A two-step estimation procedure, which ignores potential correlation between the latent state variables, suffers less from the ‘curse of dimensionality’, and it provides a reasonable alternative to the joint estimation procedure. In addition, our Monte Carlo experiments show that the two-step estimation procedure can be more efficient than the joint estimation procedure in finite samples, when there is zero or low correlation between the latent state variables.     

Variable selection and functional form uncertainty in cross-country growth regressions

Regression analyses of cross-country economic growth data are complicated by two main forms of model uncertainty: the uncertainty in selecting explanatory variables and the uncertainty in specifying the functional form of the regression function. Most discussions in the literature address these problems independently, yet a joint treatment is essential. We present a new framework that makes such a joint treatment possible, using flexible nonlinear models specified by Gaussian process priors and addressing the variable selection problem by means of Bayesian model averaging. Using this framework, we extend the linear model to allow for parameter heterogeneity of the type suggested by new growth theory, while taking into account the uncertainty in selecting explanatory variables. Controlling for variable selection uncertainty, we confirm the evidence in favor of parameter heterogeneity presented in several earlier studies. However, controlling for functional form uncertainty, we find that the effects of many of the explanatory variables identified in the literature are not robust across countries and variable selections.     

Models of military expenditure

A brief review of the extensive empirical literature devoted to explaining levels of military expenditure indicates a wide variety of unreconciled results. However, comparing the alternative models is not straightforward. This paper uses British post-war military expenditures to illustrate some of the methodological issues involved in model comparison, evaluation and selection. Starting from two published models for the UK, a systematic specification search leads to a new model, which dominates alternative specifications, passes a wide range of misspecification tests, and also fits well to French data.     

Semiparametric quantile averaging in the presence of high-dimensional predictors

The paper proposes a method for forecasting conditional quantiles. In practice, one often does not know the “true” structure of the underlying conditional quantile function, and in addition, we may have a large number of predictors. Focusing on such cases, we introduce a flexible and practical framework based on penalized high-dimensional quantile averaging. In addition to prediction, we show that the proposed method can also serve as a predictor selector. We conduct extensive simulation experiments to asses its prediction and variable selection performances for nonlinear and linear time series model designs. In terms of predictor selection, the approach tends to select the true set of predictors with minimal false positives. With respect to prediction accuracy, the method competes well even with the benchmark/oracle methods that know one or more aspects of the underlying quantile regression model. We further illustrate the merit of the proposed method by providing an application to the out-of-sample forecasting of U.S. core inflation using a large set of monthly macroeconomic variables based on FRED-MD database. The application offers several empirical findings.     

Nonparametric transfer function models

In this paper a class of nonparametric transfer function models is proposed to model nonlinear relationships between ‘input’ and ‘output’ time series. The transfer function is smooth with unknown functional forms, and the noise is assumed to be a stationary autoregressive-moving average (ARMA) process. The nonparametric transfer function is estimated jointly with the ARMA parameters. By modeling the correlation in the noise, the transfer function can be estimated more efficiently. The parsimonious ARMA structure improves the estimation efficiency in finite samples. The asymptotic properties of the estimators are investigated. The finite-sample properties are illustrated through simulations and one empirical example.