nep-for New Economics Papers
on Forecasting
Issue of 2016‒05‒21
eleven papers chosen by
Rob J Hyndman
Monash University

  1. Modelling mortality: Are we heading in the right direction? By O'Hare, Colin; Li, Youwei
  2. Dynamic Factor Models with infinite-dimensional factor space: asymptotic analysis By Mario Forni; Marc Hallin; Marco Lippi; Paolo Zaffaroni
  3. Forecast bankruptcy using a blend of clustering and MARS model - Case of US banks By Zeineb Affes; Rania Hentati-Kaffel
  4. Forecasting the Market Risk Premium with Artificial Neural Networks By Leoni Eleni Oikonomikou
  5. Does Economic Policy Uncertainty Forecast Real Housing Returns in a Panel of OECD Countries? A Bayesian Approach By Christina Christou; Rangan Gupta; Christis Hassapis
  6. Automating Analytics: Forecasting Time Series in Economics and Business By Gerunov, Anton
  7. Public news flow in intraday component models for trading activity and volatility By Adam Clements; Joanne Fuller; Vasilios Papalexiou
  8. Point process models for extreme returns: Harnessing implied volatility By R Herrera; Adam Clements
  9. Estimating point and density forecasts for the US economy with a factor-augmented vector autoregressive DSGE model By Stelios D. Bekiros; Alessia Paccagnini
  10. Forecasting time series with structural breaks with Singular Spectrum Analysis, using a general form of recurrent formula By Donya Rahmani; Saeed Heravi; Hossein Hassani; Mansi Ghodsi
  11. Quasi-Real-Time Data of the Economic Tendency Survey By Billstam, Maria; Frändén, Kristina; Samuelsson, Johan; Österholm, Pär

  1. By: O'Hare, Colin; Li, Youwei
    Abstract: Predicting life expectancy has become of upmost importance in society. Pension providers, insurance companies, government bodies and individuals in the developed world have a vested interest in understanding how long people will live for. This desire to better understand life expectancy has resulted in an explosion of stochastic mortality models many of which identify linear trends in mortality rates by time. In making use of such models for forecasting purposes we rely on the assumption that the direction of the linear trend (determined from the data used for fitting purposes) will not change in the future, recent literature has started to question this assumption. In this paper we carry out a comprehensive investigation of these types of models using male and female data from 30 countries and using the theory of structural breaks to identify changes in the extracted trends by time. We find that structural breaks are present in a substantial number of cases, that they are more prevalent in male data than in female data, that the introduction of additional period factors into the model reduces their presence, and that allowing for changes in the trend improves the fit and forecast substantially.
    Keywords: Mortality; stochastic models; structural breaks; forecasting
    JEL: C51 C52 C53 G22 G23 J11
    Date: 2016–05–16
  2. By: Mario Forni; Marc Hallin; Marco Lippi; Paolo Zaffaroni
    Abstract: Factor models, all particular cases of the Generalized Dynamic Factor Model (GDFM) introduced in Forni, Hallin, Lippi and Reichlin (2000), have become extremely popular in the theory and practice of large panels of time series data. The asymptotic properties (consistency and rates) of the corresponding estimators have been studied in Forni, Hallin, Lippi and Reichlin (2004). Those estimators, however, rely on Brillinger’s dynamic principal components, and thus involve two-sided filters, which leads to rather poor forecasting performances. No such problem arises with estimators based on standard (static) principal components, which have been dominant in this literature. On the other hand, the consistency of those static estimators requires the assumption that the space spanned by the factors has finite dimension, which severely restricts the generality afforded by the GDFM. This paper derives the asymptotic properties of a semiparametric estimator of the loadings and common shocks based on one-sided filters recently proposed by Forni, Hallin, Lippi and Zaffaroni (2015). Consistency and exact rates of convergence are obtained for this estimator, under a general class of GDFMs that does not require a finite-dimensional factor space. A Monte Carlo experiment and an empirical exercise on US macroeconomic data corroborate those theoretical results and demonstrate the excellent performance of those estimators in out-of-sample forecasting
    Keywords: High-dimensional time series. Generalized dynamic factor models. Vector processes with singular spectral density. One-sided representations of dynamic factor models. Consistency and rates
    JEL: C0 C01 E0
    Date: 2015–09
  3. By: Zeineb Affes (CES - Centre d'économie de la Sorbonne - UP1 - Université Panthéon-Sorbonne - CNRS - Centre National de la Recherche Scientifique); Rania Hentati-Kaffel (CES - Centre d'économie de la Sorbonne - UP1 - Université Panthéon-Sorbonne - CNRS - Centre National de la Recherche Scientifique)
    Abstract: In this paper, we compare the performance of two non-parametric methods of classification, Regression Trees (CART) and the newly Multivariate Adaptive Regression Splines (MARS) models, in forecasting bankruptcy. Models are implemented on a large universe of US banks over a complete market cycle and running under a K-Fold Cross validation. A hybrid model which combines K-means clustering and MARS is tested as well. Our findings highlight that i) Either in training or testing sample, MARS provides, in average, better correct classification rate than CART model, ii) Hybrid approach significantly enhances the classification accuracy rate for both the training and the testing samples, iii) MARS prediction underperforms when the misclassification rate is adopted as a criteria, iv) Results proves that Non-parametric models are more suitable for bank failure prediction than the corresponding Logit model.
    Keywords: Bankruptcy prediction,MARS,CART,K-means,Early-Warning System
    Date: 2016–03
  4. By: Leoni Eleni Oikonomikou (Georg-August University Göttingen)
    Abstract: This paper aims to forecast the Market Risk premium (MRP) in the US stock market by applying machine learning techniques, namely the Multilayer Perceptron Network (MLP), the Elman Network (EN) and the Higher Order Neural Network (HONN). Furthermore, Univariate ARMA and Exponential Smoothing models are also tested. The Market Risk Premium is defined as the historical differential between the return of the benchmark stock index over a short-term interest rate. Data are taken in daily frequency from January 2007 through December 2014. All these models outperform a Naive benchmark model. The Elman network outperforms all the other models during the insample period, whereas the MLP network provides superior results in the out-of-sample period. The contribution of this paper to the existing literature is twofold. First, it is the first study that attempts to forecast the Market Risk Premium in a daily basis using Artificial Neural Networks (ANNs). Second, it is not based on a theoretical model but is mainly data driven. The chosen calculation approach fits quite well with the characteristics of ANNs. The forecasting model is tested with data from the US stock market. The proposed model-based forecasting method aims to capture patterns in the data that improve the forecasting accuracy of the Market Risk Premium in the tested market and indicates potential key metrics for investment and trading purposes for short time horizons.
    Keywords: nonlinear models; forecasting performance metrics; market risk premium; US equity market
    JEL: C45 C52 G15 G17
    Date: 2016–04–14
  5. By: Christina Christou (School of Economics and Management, Open University of Cyprus, 2252, Latsia, Cyprus); Rangan Gupta (Department of Economics, University of Pretoria); Christis Hassapis (Department of Economics, University of Cyprus, P.O. Box 20537, CY-1678 Nicosia, Cyprus)
    Abstract: This paper investigates whether the news-based measure of economic policy uncertainty (EPU) could help in forecasting the real housing returns in ten (Canada, France, Germany, Italy, Japan, The Netherlands, South Korea, Spain, United Kingdom, and United States of America) Organization for Economic Co-operation and Development (OECD) countries. We analyze the quarterly out-of-sample period of 2008:Q2-2014:Q4, given an in-sample period of 2003:Q1- 2008:1Q1, using time series and panel data-based vector autoregressive models, with the latter allowing for heterogeneity, and static and dynamic interdependence. It is found that regardless of the forecasting model considered, EPU is useful for forecasting real housing returns. Our results show that, panel data models, especially the Bayesian variants which allow for parameter shrinkage, consistently beat time series autoregressive models suggesting the importance of pooling information when trying to forecast real housing returns.
    Keywords: Real Housing Returns, Economic Policy Uncertainty, OECD Countries, Panel Vector Autoregressions
    JEL: C33 C53 R31
    Date: 2016–04
  6. By: Gerunov, Anton
    Abstract: With the growing ability of organizations in the public and private sector to collect large volumes of real-time data, the mounting pile of information presents specific challenges for storage, processing, and analysis. Many organizations do need data analysis for the purposes of planning and logistics. Likewise, governments and regulators will need analysis to support policy-making, implementation and controlling. All this leads to the importance of being able to generate large scale analytics under (sometimes severe) resource constraints. This paper investigates a possible solution – automating analytics with a special focus on forecasting time series. Such approach has the benefit of being able to produce scalable forecasting of thousands of variables with relatively high accuracy for a short period of time and few resources. We first review the literature on time series forecasting with a particular focus on the M, M-2, and M-3 forecasting competition and outline a few major conclusions supported across different empirical studies. The paper then proceeds to explore the typical structure of a time-series variables using Bulgarian GDP growth and show how the ARIMA modeling with a seasonal component can be used to fit economic data of this class. We also review some major approaches to automating forecasting and outline the benefits of selecting the optimal model from a large set of ARIMA alternatives using an information criterion. A possible approach to fit an automated forecasting algorithm on four crucial economic time series from the Bulgarian economy is demonstrated. We use data on GDP growth, inflation, unemployment, and interest rates and fit a large number of possible models. The best ones are selected by taking recourse to the Akaike Information Criterion. The optimal ARIMA models are studied and commented. Forecast accuracy metrics are presented and a few major conclusions and possible model applications are outlined. The paper concludes with directions for further research.
    Keywords: Automated analytics, forecasting, time series, ARIMA, business forecasting
    JEL: C22 C53 E37
    Date: 2016–04
  7. By: Adam Clements (QUT); Joanne Fuller (QUT); Vasilios Papalexiou
    Abstract: Understanding the determinants of, and forecasting asset return volatility are crucial issues in many financial applications. Many earlier studies have considered the impact of trading activity and news arrivals on volatility. This paper develops a range of intraday component models for volatility and order flow that include the impact of news arrivals. Estimates of the conditional mean of order flow, taking into account news flow are included in models ofvolatility providing a superior in-sample fit. At a 1-minute frequency, it is found that first generating forecasts of order flow which are then included in forecasts of volatility leads to superior day-ahead forecasts of volatility. While including overnight news arrivals directly into models for volatility improves in-sample fit, this approach produces inferior forecasts.
    Keywords: Volatility; Order flow; News; Dynamic conditional score; forecasting
    JEL: C22 G00
    Date: 2015–08–26
  8. By: R Herrera; Adam Clements (QUT)
    Abstract: Forecasting the risk of extreme losses is an important issue in the management of financial risk. There has been a great deal of research examining how option implied volatilities (IV) can be used to forecasts asset return volatility. However, the impact of IV in the context of predicting extreme risk has received relatively little attention. The role of IV is considered within a range of models beginning with the traditional GARCH based approach. Furthermore, a number of novel point process models for forecasting extreme risk are proposed in this paper. Univariate models where IV is included as an exogenous variable are considered along with a novel bivariate approach where movements in IV are treated as another point process. It is found that in the context of forecasting Value-at-Risk, the bivariate models produce the most accurate forecasts across a wide range of scenarios.
    Keywords: Implied volatility, Hawkes process, Peaks over threshold, Point process, Extreme events
    JEL: C14 C53
    Date: 2015–05–06
  9. By: Stelios D. Bekiros; Alessia Paccagnini
    Abstract: Although policymakers and practitioners are particularly interested in dynamic stochastic general equilibrium (DSGE) models, these are typically too stylized to be applied directly to the data and often yield weak prediction results. Very recently, hybrid DSGE models have become popular for dealing with some of the model misspecifications. Major advances in estimation methodology could allow these models to outperform well-known time series models and effectively deal with more complex real-world problems as richer sources of data become available. In this study we introduce a Bayesian approach to estimate a novel factor augmented DSGE model that extends the model of Consolo et al. [Consolo, A., Favero, C.A., and Paccagnini, A., 2009. On the Statistical Identification of DSGE Models. Journal of Econometrics, 150, 99–115]. We perform a comparative predictive evaluation of point and density forecasts for many different specifications of estimated DSGE models and various classes of VAR models, using datasets from the US economy including real-time data. Simple and hybrid DSGE models are implemented, such as DSGE-VAR and tested against standard, Bayesian and factor augmented VARs. The results can be useful for macro-forecasting and monetary policy analysis.
    Keywords: Density forecasting; Marginal data density; DSGE-FAVAR; Real-Time data
    JEL: C32 C11 C15 C53 D58
    Date: 2014–10
  10. By: Donya Rahmani; Saeed Heravi; Hossein Hassani; Mansi Ghodsi
    Abstract: This study extends and evaluates the forecasting performance of the Singular Spectrum Analysis (SSA) technique using a general non-linear form for the re- current formula. In this study, we consider 24 series measuring the monthly seasonally adjusted industrial production of important sectors of the German, French and UK economies. This is tested by comparing the performance of the new proposed model with basic SSA and the SSA bootstrap forecasting, especially when there is evidence of structural breaks in both in-sample and out-of-sample periods. According to root mean-square error (RMSE), SSA using the general recursive formula outperforms both the SSA and the bootstrap forecasting at horizons of up to a year. We found no significant difference in predicting the direction of change between these methods. Therefore, it is suggested that the SSA model with the general recurrent formula should be chosen by users in the case of structural breaks in the series.
    Date: 2016–05
  11. By: Billstam, Maria (National Institute of Economic Research); Frändén, Kristina (National Institute of Economic Research); Samuelsson, Johan (National Institute of Economic Research); Österholm, Pär (National Institute of Economic Research)
    Abstract: Survey data from businesses and households are widely used for fore-casting and economic analysis. In Sweden, the most important survey of this kind is the Economic Tendency Survey of the National Institute of Economic Research. A shortcoming with this survey is that real-time data of it largely are unavailable. In this paper, we describe how two quasi-real-time data sets of this survey have been constructed and made publicly available – one monthly and one quarterly. The data sets consist of monthly/quarterly vintages of the most important series of the survey, including the main confidence indicators. A natural usage of these data sets is evaluations of model-based forecasts and nowcasts. We illustrate this with an application to Swedish GDP growth. This shows that all models based on indicators from the Economic Tendency Survey, except the one relying on the confidence indicator for the construction industry, have higher forecast precision than the benchmark models.
    Keywords: Data revisions; Nowcasting
    JEL: C83 E17
    Date: 2016–04–26

This nep-for issue is ©2016 by Rob J Hyndman. It is provided as is without any express or implied warranty. It may be freely redistributed in whole or in part for any purpose. If distributed in part, please include this notice.
General information on the NEP project can be found at For comments please write to the director of NEP, Marco Novarese at <>. Put “NEP” in the subject, otherwise your mail may be rejected.
NEP’s infrastructure is sponsored by the School of Economics and Finance of Massey University in New Zealand.