nep-for New Economics Papers
on Forecasting
Issue of 2019‒05‒06
six papers chosen by
Rob J Hyndman
Monash University

  1. Combining Multivariate Volatility Forecasts using Weighted Losses By A Clements; M Doolan
  2. A Practical Guide to Harnessing the HAR Volatility Model By A Clements; D Preve
  3. Forecasting Realized Volatility of Russian stocks using Google Trends and Implied Volatility By Bazhenov, Timofey; Fantazzini, Dean
  4. Volatility forecasts for the RTS stock index: option-implied volatility versus alternative methods By Guglielmo Maria Caporale; Daria Teterkina
  5. Curriculum Learning in Deep Neural Networks for Financial Forecasting By Allison Koenecke; Amita Gajewar
  6. Modeling of Economic and Financial Conditions for Nowcasting and Forecasting Recessions: A Unified Approach By Cem Cakmakli; Hamza Demircan; Sumru Altug

  1. By: A Clements; M Doolan
    Abstract: The ability to improve out-of-sample forecasting performance by combining forecasts is well established in the literature. This paper advances this literature in the area of multivariate volatility forecasts by developing two combination weighting schemes that are capable of placing varying emphasis on losses within the combination estimation period. A comprehensive empirical analysis of the out-of-sample forecast performance across varying dimensions, loss functions, sub-samples and forecast horizons show that new approaches significantly outperform their counterparts in terms of statistical accuracy. Within the financial applications considered, significant benefits from combination forecasts relative to the individual candidate models are observed. Although the more sophisticated combination approaches consistently rank higher relative to the equally weighted approach, their performance is statistically indistinguishable given the relatively low power of these loss functions. Finally, within the applications, further analysis highlights how combination forecasts dramatically reduce the variability in the parameter of interest, namely the portfolio weight or beta.
    Keywords: Multivariate volatility, combination forecasts, forecast evaluation, model confidence set
    JEL: C22 G00
    Date: 2018–12–11
  2. By: A Clements; D Preve
    Abstract: The standard heterogeneous autoregressive (HAR) model is perhaps the most popular benchmark model for forecasting return volatility. It is often estimated using raw realized variance (RV) and ordinary least squares (OLS). However, given the stylized facts of RV and wellknown properties of OLS, this combination should be far from ideal. One goal of this paper is to investigate how the predictive accuracy of the HAR model depends on the choice of estimator, transformation, and forecasting scheme made by the market practitioner. Another goal is to examine the effect of replacing its high-frequency data based volatility proxy (RV) with a proxy based on free and publicly available low-frequency data (logarithmic range). In an out-of-sample study, covering three major stock market indices over 16 years, it is found that simple remedies systematically outperform not only standard HAR but also state of the art HARQ forecasts, and that HAR models using logarithmic range can often produce forecasts of similar quality to those based on RV.
    Keywords: Volatility forecasting; Realized variance; HAR model; HARQ model; Robust regression; Box-Cox transformation; Forecast comparisons; QLIKE loss; Model confidence set
    JEL: C22 C51 C52 C53 C58
    Date: 2019–04–12
  3. By: Bazhenov, Timofey; Fantazzini, Dean
    Abstract: This work proposes to forecast the Realized Volatility (RV) and the Value-at-Risk (VaR) of the most liquid Russian stocks using GARCH, ARFIMA and HAR models, including both the implied volatility computed from options prices and Google Trends data. The in-sample analysis showed that only the implied volatility had a significant effect on the realized volatility across most stocks and estimated models, whereas Google Trends did not have any significant effect. The out-of-sample analysis highlighted that models including the implied volatility improved their forecasting performances, whereas models including internet search activity worsened their performances in several cases. Moreover, simple HAR and ARFIMA models without additional regressors often reported the best forecasts for the daily realized volatility and for the daily Value-at-Risk at the 1% probability level, thus showing that efficiency gains more than compensate any possible model misspecifications and parameters biases. Our empirical evidence shows that, in the case of Russian stocks, Google Trends does not capture any additional information already included in the implied volatility.
    Keywords: Forecasting; Realized Volatility; Value-at-Risk; Implied Volatility; Google Trends; GARCH; ARFIMA; HAR;
    JEL: C22 C51 C53 G17 G32
    Date: 2019–04
  4. By: Guglielmo Maria Caporale; Daria Teterkina
    Abstract: This paper compares volatility forecasts for the RTS Index (the main index for the Russian stock market) generated by alternative models, specifically option-implied volatility forecasts based on the Black-Scholes model, ARCH/GARCH-type model forecasts, and forecasts combining those two using a mixing strategy based either on a simple average or a weighted average with the weights being determined according to two different criteria (either minimizing the errors or maximizing the information content). Various forecasting performance tests are carried out which suggest that both implied volatility and combination methods using a simple average outperform ARCH/GARCH-type models in terms of forecasting accuracy.
    Keywords: option-implied volatility, ARCH-type models, mixed strategies
    JEL: C22 G12
    Date: 2019
  5. By: Allison Koenecke; Amita Gajewar
    Abstract: For any financial organization, computing accurate quarterly forecasts for various products is one of the most critical operations. As the granularity at which forecasts are needed increases, traditional statistical time series models may not scale well. We apply deep neural networks in the forecasting domain by experimenting with techniques from Natural Language Processing (Encoder-Decoder LSTMs) and Computer Vision (Dilated CNNs), as well as incorporating transfer learning. A novel contribution of this paper is the application of curriculum learning to neural network models built for time series forecasting. We illustrate the performance of our models using Microsoft's revenue data corresponding to Enterprise, and Small, Medium & Corporate products, spanning approximately 60 regions across the globe for 8 different business segments, and totaling in the order of tens of billions of USD. We compare our models' performance to the ensemble model of traditional statistics and machine learning techniques currently used by Microsoft Finance. With this in-production model as a baseline, our experiments yield an approximately 30% improvement in overall accuracy on test data. We find that our curriculum learning LSTM-based model performs best, showing that it is reasonable to implement our proposed methods without overfitting on medium-sized data.
    Date: 2019–04
  6. By: Cem Cakmakli (Koc University); Hamza Demircan (Koc University); Sumru Altug (American University of Beirut, CEPR)
    Abstract: In this paper, we propose a method for jointly estimating indexes of economic and financial conditions by exploiting the intertemporal link between their cyclical behavior. This method combines a dynamic factor model for the joint modeling of economic and financial variables with mixed frequencies together with a tailored Markov regime switching specification for capturing their cyclical behavior. It allows for imperfect synchronization between the cycles in economic and financial conditions/factors by explicitly estimating the phase shifts between their cyclical regimes. We examine the efficacy of the model for predicting cyclical activity in a key emerging economy, namely, Turkey, by making use of a mixed frequency ragged-edge data set. A comparison of our framework with more conventional cases imposing common cyclical dynamics as well as independent cyclical dynamics for the economic and financial indicators reveals that the proposed specification provides precise estimates of indexes of economic and financial activity together with accurate and timely recession probabilities. Recession probabilities estimated using the available data in the first week of November 2018 indicate that Turkey entered a recession that is still ongoing starting from August 2018. We further conduct a recursive real-time exercise of nowcasting and forecasting business cycle turning points. The results show evidence for the superior predictive power of our specification by signaling oncoming recessions (expansions) as early as 3.6 (3.3) months ahead of the actual realization.
    Keywords: Financial conditions index; Coincident economic index; Dynamic factor model; Markov switching; Imperfect synchronization; Bayesian inference.
    JEL: C11 C32 C38 E37
    Date: 2019–04

This nep-for issue is ©2019 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.