nep-for New Economics Papers
on Forecasting
Issue of 2021‒07‒12
five papers chosen by
Rob J Hyndman
Monash University

  1. Forecasting Canadian GDP Growth with Machine Learning By Shafiullah Qureshi; Ba Chu; Fanny S. Demers
  2. Comparison of the accuracy in VaR forecasting for commodities using different methods of combining forecasts By Szymon Lis; Marcin Chlebus
  3. Quantifying time-varying forecast uncertainty and risk for the real price of oil By Knut Are Aastveit; Jamie Cross; Herman K. Djik
  4. Forecasting the intra-day effective bid ask spread by combining density forecasts By Malick Fall; Waël Louhichi; Jean Viviani
  5. The incremental information in the yield curve about future interest rate risk By Bent Jesper Christensen; Mads Markvart Kjær; Bezirgen Veliyev

  1. By: Shafiullah Qureshi (Department of Economics, Carleton University); Ba Chu (Department of Economics, Carleton University); Fanny S. Demers (Department of Economics, Carleton University)
    Abstract: This paper applies state-of-the-art machine learning (ML) algorithms to forecast monthly real GDP growth in Canada by using both Google Trends (GT) data and official macroeconomic data (which are available ahead of the release of GDP data by Statistics Canada). We show that we can forecast real GDP growth accurately ahead of the release of GDP figures by using GT and official data (such as employment) as predictors. We first pre-select features by applying up-to-date techniques, namely, XGBoost’s variable importance score, and a recent variable-screening procedure for time series data, namely, PDC-SIS+. These pre-selected features are then used to build advanced ML models for forecasting real GDP growth, by employing tree-based ensemble algorithms, such as XGBoost, LightGBM, Random Forest, and GBM. We provide empirical evidence that the variables pre-selected by either PDC-SIS+ or the XGBoost’s variable importance score can have a superior forecasting ability. We find that the pre-selected GT data features perform as well as the pre-selected official data features with respect to short-term forecasting ability, while the pre-selected official data features are superior with respect to long-term forecasting ability. We also find that (1) the ML algorithms we employ often perform better with a large sample than with a small sample, even when the small sample has a larger set of predictors; and (2) the Random Forest (that often produces nonlinear models to capture nonlinear patterns in the data) tends to under-perform a standard autoregressive model in several cases while there is no clear evidence that the XGBoost and the LightGBM can always outperform each other.
    Date: 2021–05–17
  2. By: Szymon Lis (Faculty of Economic Sciences, University of Warsaw); Marcin Chlebus (Faculty of Economic Sciences, University of Warsaw)
    Abstract: No model dominates existing VaR forecasting comparisons. This problem may be solved by combine forecasts. This study investigates the daily volatility forecasting for commodities (gold, silver, oil, gas, copper) from 2000-2020 and identifies the source of performance improvements between individual GARCH models and combining forecasts methods (mean, the lowest, the highest, CQOM, quantile regression with the elastic net or LASSO regularization, random forests, gradient boosting, neural network) through the MCS. Results indicate that individual models achieve more accurate VaR forecasts for the confidence level of 0.975, but combined forecasts are more precise for 0.99. In most cases simple combining methods (mean or the lowest VaR) are the best. Such evidence demonstrates that combining forecasts is important to get better results from the existing models. The study shows that combining the forecasts allows for more accurate VaR forecasting, although it’s difficult to find accurate, complex methods.
    Keywords: Combining forecasts, Econometric models, Finance, Financial markets, GARCH models, Neural networks, Regression, Time series, Risk, Value-at-Risk, Machine learning, Model Confidence Set
    JEL: C51 C52 C53 G32 Q01
    Date: 2021
  3. By: Knut Are Aastveit; Jamie Cross; Herman K. Djik
    Abstract: We propose a novel and numerically efficient quantification approach to forecast uncertainty of the real price of oil using a combination of probabilistic individual model forecasts. Our combination method extends earlier approaches that have been applied to oil price forecasting, by allowing for sequentially updating of time-varying combination weights, estimation of time-varying forecast biases and facets of miscalibration of individual forecast densities and time-varying inter-dependencies among models. To illustrate the usefulness of the method, we present an extensive set of empirical results about time-varying forecast uncertainty and risk for the real price of oil over the period 1974-2018. We show that the combination approach systematically outperforms commonly used benchmark models and combination approaches, both in terms of point and density forecasts. The dynamic patterns of the estimated individual model weights are highly time-varying, reflecting a large time variation in the relative performance of the various individual models. The combination approach has built-in diagnostic information measures about forecast inaccuracy and/or model set incompleteness, which provide clear signals of model incompleteness during three crisis periods. To highlight that our approach also can be useful for policy analysis, we present a basic analysis of profit-loss and hedging against price risk.
    Keywords: Oil price, Forecast density combination, Bayesian forecasting, Instabilities, Model uncertainty
    Date: 2021–06
  4. By: Malick Fall (CREM - Centre de recherche en économie et management - UNICAEN - Université de Caen Normandie - NU - Normandie Université - UR1 - Université de Rennes 1 - UNIV-RENNES - Université de Rennes - CNRS - Centre National de la Recherche Scientifique); Waël Louhichi (ESSCA Research Lab - ESSCA - Groupe ESSCA); Jean Viviani (CREM - Centre de recherche en économie et management - UNICAEN - Université de Caen Normandie - NU - Normandie Université - UR1 - Université de Rennes 1 - UNIV-RENNES - Université de Rennes - CNRS - Centre National de la Recherche Scientifique)
    Abstract: The bid-ask spread refers to the tightness dimension of liquidity and can be used as a proxy for transaction costs. Despite the importance of the bid-ask spread in the financial literature, few studies have investigated its forecastability. We propose a new methodology to predict the bid ask spread by combining density forecasts of two types of models: Multiplicative Errors Models and ARMA-GARCH models. Our method is employed to predict the effective intra-day bid-ask spread series of all shares pertaining to the CAC40 index. Using a one-step-ahead out-of-sample framework, we resort on the Model Confidence Set procedure of Hansen et al. (2004) to classify models and we found that the proposed model appears to beat all the benchmark specifications.
    Keywords: Effective bid-ask spread,High-Frequency,Multiplicative Errors Models,Forecasting
    Date: 2021
  5. By: Bent Jesper Christensen (Aarhus University and CREATES and the Dale T. Mortensen Center); Mads Markvart Kjær (Aarhus University and CREATES); Bezirgen Veliyev (Aarhus University and CREATES and the Danish Finance Institute)
    Abstract: Using high-frequency intraday futures prices to measure yield volatility at selected maturities, we find that daily yield curves carry incremental information about future interest rate risk at the long end, relative to that contained in the time series of historical volatilities. Some of the information in the yield curves is not captured by standard affine models. At the short end, time series based forecasts outperform yield curve based forecasts. Both provide utility to a risk averse investor in longerterm instruments, not in short, relative to a random walk. Our results point to the existence of an unspanned volatility factor.
    Keywords: Term structure models, Volatility, Forecasting, Kalman filtering, Yield curve
    JEL: C58 E43 G12
    Date: 2021–07–01

This nep-for issue is ©2021 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.