nep-for New Economics Papers
on Forecasting
Issue of 2020‒11‒16
ten papers chosen by
Rob J Hyndman
Monash University

  1. Stock Price Prediction Using CNN and LSTM-Based Deep Learning Models By Sidra Mehtab; Jaydip Sen
  2. Learning from Forecast Errors: A New Approach to Forecast Combination By Tae-Hwy Lee; Ekaterina Seregina
  3. Two out-of-sample forecasting models of the equity premium By de Oliveira Souza, Thiago
  4. News Media vs. FRED-MD for Macroeconomic Forecasting By Jon Ellingsen; Vegard H. Larsen; Leif Anders Thorsrud
  5. Recurrent Conditional Heteroskedasticity By T. -N. Nguyen; M. -N. Tran; R. Kohn
  6. Modelling Returns in US Housing Prices – You’re the One for Me, Fat Tails By Kiss, Tamás; Nguyen, Hoang; Österholm, Pär
  7. Forecasting Quarterly Brazilian GDP: Univariate Models Approach By Kleyton Vieira Sales da Costa; Felipe Leite Coelho da Silva; Josiane da Silva Cordeiro Coelho
  8. Comparison of ARIMA, ETS, NNAR and hybrid models to forecast the second wave of COVID-19 hospitalizations in Italy By Perone, G.
  9. Uncertainty due to Infectious Diseases and Forecastability of the Realized Variance of US REITs: A Note By Matteo Bonato; Oguzhan Cepni; Rangan Gupta; Christian Pierdzioch
  10. Do Oil-Price Shocks Predict the Realized Variance of U.S. REITs? By Matteo Bonato; Rangan Gupta; Christian Pierdzioch

  1. By: Sidra Mehtab; Jaydip Sen
    Abstract: Designing robust and accurate predictive models for stock price prediction has been an active area of research for a long time. While on one side, the supporters of the efficient market hypothesis claim that it is impossible to forecast stock prices accurately, many researchers believe otherwise. There exist propositions in the literature that have demonstrated that if properly designed and optimized, predictive models can very accurately and reliably predict future values of stock prices. This paper presents a suite of deep learning based models for stock price prediction. We use the historical records of the NIFTY 50 index listed in the National Stock Exchange of India, during the period from December 29, 2008 to July 31, 2020, for training and testing the models. Our proposition includes two regression models built on convolutional neural networks and three long and short term memory network based predictive models. To forecast the open values of the NIFTY 50 index records, we adopted a multi step prediction technique with walk forward validation. In this approach, the open values of the NIFTY 50 index are predicted on a time horizon of one week, and once a week is over, the actual index values are included in the training set before the model is trained again, and the forecasts for the next week are made. We present detailed results on the forecasting accuracies for all our proposed models. The results show that while all the models are very accurate in forecasting the NIFTY 50 open values, the univariate encoder decoder convolutional LSTM with the previous two weeks data as the input is the most accurate model. On the other hand, a univariate CNN model with previous one week data as the input is found to be the fastest model in terms of its execution speed.
    Date: 2020–10
  2. By: Tae-Hwy Lee (Department of Economics, University of California Riverside); Ekaterina Seregina (University of California Riverside)
    Abstract: This paper studies forecast combination (as an expert system) using the precision matrix estimation of forecast errors when the latter admit the approximate factor model. This approach incorporates the facts that experts often use common sets of information and hence they tend to make common mistakes. This premise is evidenced in many empirical results. For example, the European Central Bank's Survey of Professional Forecasters on Euro-area real GDP growth demonstrates that the professional forecasters tend to jointly understate or overstate GDP growth. Motivated by this stylized fact, we develop a novel framework which exploits the factor structure of forecast errors and the sparsity in the precision matrix of the idiosyncratic components of the forecast errors. The proposed algorithm is called Factor Graphical Model (FGM). Our approach overcomes the challenge of obtaining the forecasts that contain unique information, which was shown to be necessary to achieve a "winning" forecast combination. In simulation, we demonstrate the merits of the FGM in comparison with the equal-weighted forecasts and the standard graphical methods in the literature. An empirical application to forecasting macroeconomic time series in big data environment highlights the advantage of the FGM approach in comparison with the existing methods of forecast combination.
    Keywords: High-dimensionality, Graphical Lasso, Approximate Factor Model, Nodewise Regression, Precision Matrix
    JEL: C13 C38 C55
    Date: 2020–09
  3. By: de Oliveira Souza, Thiago (Department of Business and Economics)
    Abstract: I derive two valid forecasting models of the equity premium in monthly frequency, based on little more than no-arbitrage: A “predictability timing” version of partial least squares, given that predictability is theoretically timevarying; and a least squares model with realized market premiums in monthly frequency as the regressor, since realized returns are theoretically correlated to risk and to the price of risk. This evidence is consistent with the instability inherent to monthly equity premium forecasts based on standard partial least squares and disaggregated book-to-markets as regressors, and with the fact that taking one extra lag of book-to-markets in predictive return regressions improves the estimates.
    Keywords: Predictability; out-of-sample; equity premium; disaggregated book-to-markets
    JEL: G11 G12 G14
    Date: 2020–10–27
  4. By: Jon Ellingsen; Vegard H. Larsen; Leif Anders Thorsrud
    Abstract: Using a unique dataset of 22.5 million news articles from the Dow Jones Newswires Archive, we perform an in depth real-time out-of-sample forecasting comparison study with one of the most widely used data sets in the newer forecasting literature, namely the FRED-MD dataset. Focusing on U.S. GDP, consumption and investment growth, our results suggest that the news data contains information not captured by the hard economic indicators, and that the news-based data are particularly informative for forecasting consumption developments.
    Keywords: forecasting, real-time, machine learning, news, text data
    JEL: C53 C55 E27 E37
    Date: 2020
  5. By: T. -N. Nguyen; M. -N. Tran; R. Kohn
    Abstract: We propose a new class of financial volatility models, which we call the REcurrent Conditional Heteroskedastic (RECH) models, to improve both the in-sample analysis and out-of-sample forecast performance of the traditional conditional heteroskedastic models. In particular, we incorporate auxiliary deterministic processes, governed by recurrent neural networks, into the conditional variance of the traditional conditional heteroskedastic models, e.g. the GARCH-type models, to flexibly capture the dynamics of the underlying volatility. The RECH models can detect interesting effects in financial volatility overlooked by the existing conditional heteroskedastic models such as the GARCH (Bollerslev, 1986), GJR (Glosten et al., 1993) and EGARCH (Nelson, 1991). The new models often have good out-of-sample forecasts while still explain well the stylized facts of financial volatility by retaining the well-established structures of the econometric GARCH-type models. These properties are illustrated through simulation studies and applications to four real stock index datasets. An user-friendly software package together with the examples reported in the paper are available at
    Date: 2020–10
  6. By: Kiss, Tamás (Örebro University School of Business); Nguyen, Hoang (Örebro University School of Business); Österholm, Pär (Örebro University School of Business)
    Abstract: In this paper, we analyse the heavy-tailed behaviour in the dynamics of housing-price returns in the United States. We investigate the sources of heavy tails by estimating autoregressive models in which innovations can be subject to GARCH effects and/or non-Gaussianity. Using monthly data ranging from January 1954 to September 2019, the properties of the models are assessed both within- and out-of-sample. We find strong evidence in favour of modelling both GARCH effects and non-Gaussianity. Accounting for these properties improves within-sample performance as well as point and density forecasts.
    Keywords: Non-Gaussianity; GARCH; Density forecasts; Probability integral transform
    JEL: C22 C52 E44 E47 G17
    Date: 2020–10–29
  7. By: Kleyton Vieira Sales da Costa; Felipe Leite Coelho da Silva; Josiane da Silva Cordeiro Coelho
    Abstract: Gross domestic product (GDP) is an important economic indicator that aggregates useful information to assist economic agents and policymakers in their decision-making process. In this context, GDP forecasting becomes a powerful decision optimization tool in several areas. In order to contribute in this direction, we investigated the efficiency of classical time series models and the class of state-space models, applied to Brazilian gross domestic product. The models used were: a Seasonal Autoregressive Integrated Moving Average (SARIMA) and a Holt-Winters method, which are classical time series models; and the dynamic linear model, a state-space model. Based on statistical metrics of model comparison, the dynamic linear model presented the best forecasting model and fit performance for the analyzed period, also incorporating the growth rate structure significantly.
    Date: 2020–10
  8. By: Perone, G.
    Abstract: Coronavirus disease (COVID-19) is a severe ongoing novel pandemic that has emerged in Wuhan, China, in December 2019. As of October 13, the outbreak has spread rapidly across the world, affecting over 38 million people, and causing over 1 million deaths. In this article, I analysed several time series forecasting methods to predict the spread of COVID-19 second wave in Italy, over the period after October 13, 2020. I used an autoregressive model (ARIMA), an exponential smoothing state space model (ETS), a neural network autoregression model (NNAR), and the following hybrid combinations of them: ARIMA-ETS, ARIMA-NNAR, ETS-NNAR, and ARIMA-ETS-NNAR. About the data, I forecasted the number of patients hospitalized with mild symptoms, and in intensive care units (ICU). The data refer to the period February 21, 2020– October 13, 2020 and are extracted from the website of the Italian Ministry of Health ( The results show that i) the hybrid models, except for ARIMA-ETS, are better at capturing the linear and non-linear epidemic patterns, by outperforming the respective single models; and ii) the number of COVID-19-related hospitalized with mild symptoms and in ICU will rapidly increase in the next weeks, by reaching the peak in about 50-60 days, i.e. in mid-December 2020, at least. To tackle the upcoming COVID-19 second wave it is necessary to enhance social distancing, hire healthcare workers and implement sufficient hospital facilities, protective equipment, and ordinary and intensive care beds.
    Keywords: COVID-19; outbreak; second wave; Italy; hybrid forecasting models; ARIMA; ETS; NNAR.
    JEL: C22 C53 I18
    Date: 2020–11
  9. By: Matteo Bonato (Department of Economics and Econometrics, University of Johannesburg, Auckland Park, South Africa; IPAG Business School, 184 Boulevard Saint-Germain, 75006 Paris, France); Oguzhan Cepni (Copenhagen Business School, Department of Economics, Porcelænshaven 16A, Frederiksberg DK-2000, Denmark; Central Bank of the Republic of Turkey, Haci Bayram Mah. Istiklal Cad. No:10 06050, Ankara, Turkey); Rangan Gupta (Department of Economics, University of Pretoria, Pretoria, 0002, South Africa); Christian Pierdzioch (Department of Economics, Helmut Schmidt University, Holstenhofweg 85, P.O.B. 700822, 22008 Hamburg, Germany)
    Abstract: We examine the forecasting power of a daily newspaper-based index of uncertainty associated with infectious diseases (EMVID) for Real Estate Investment Trusts (REITs) realized market variance of the United States (US) via the heterogeneous autoregressive realized volatility (HAR-RV) model. Our results show that the EMVID index improves the forecast accuracy of realized variance of REITs at short-, medium-, and long-run horizons in a statistically significant manner, with the result being robust to the inclusion of additional controls (leverage, realized jumps, skewness, and kurtosis) capturing extreme market movements, and also carries over to ten sub-sectors of the US REITs market. Our results have important portfolio implications for investors during the current period of unprecedented levels of uncertainty resulting from the outbreak of COVID-19.
    Keywords: Uncertainty, Infectious diseases, REITs, Realized variance, Forecasting
    JEL: C22 C53 G10
    Date: 2020–10
  10. By: Matteo Bonato (Department of Economics and Econometrics, University of Johannesburg, Auckland Park, South Africa; IPAG Business School, 184 Boulevard Saint-Germain, 75006 Paris, France; Copenhagen Business School, Department of Economics, Porcelænshaven 16A, Frederiksberg DK-2000, Denmark; Central Bank of the Republic of Turkey, Haci Bayram Mah. Istiklal Cad. No:10 06050, Ankara, Turkey); Rangan Gupta (Department of Economics, University of Pretoria, Pretoria, 0002, South Africa); Christian Pierdzioch (Department of Economics, Helmut Schmidt University, Holstenhofweg 85, P.O.B. 700822, 22008 Hamburg, Germany)
    Abstract: We examine, using aggregate and sectoral U.S. data for the period 2008-2020, the predictive power of disentangled oil-price shocks for Real Estate Investment Trusts (REITs) realized market variance via the heterogeneous auto-regressive realized variance (HAR-RV) model. In-sample tests show that demand and financial-market risk shocks contribute to a larger extent to the overall fit of the model than supply shocks, where the in-sample transmission of the impact of the shocks mainly operates through their significant effects on realized upward (“good†) variance. Out-of-sample tests corroborate the significant predictive value of demand and risk shocks for realized variance and its upward counterpart at a short, medium, and long forecast horizon, for various recursive-estimation windows, for realized volatility (that is, the square root of realized variance), for a shorter sub-sample period that excludes the recent phase of exceptionally intense oil-market turbulence, and for an extended benchmark model that features realized higher-order moments, realized jumps, and a leverage effect as control variables.
    Keywords: Oil price; Shocks, REITs; Realized variance; Forecasting
    JEL: C22 C53 G10 Q02
    Date: 2020–11

This nep-for issue is ©2020 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.