nep-for New Economics Papers
on Forecasting
Issue of 2022‒12‒12
four papers chosen by
Rob J Hyndman
Monash University

  1. Forecasting the Stability and Growth Pact compliance using Machine Learning By Kea Baret; Amelie Barbier-Gauchard; Theophilos Papadimitriou
  2. Density and Risk Prediction with Non-Gaussian COMFORT Models By Marc S. Paolella; Pawel Polak
  3. Predicting football outcomes from Spanish league using machine learning models By Michał Lewandowski; Marcin Chlebus
  4. Robust estimation for Threshold Autoregressive Moving-Average models By Greta Goracci; Davide Ferrari; Simone Giannerini; Francesco ravazzolo

  1. By: Kea Baret (University of Strasbourg); Amelie Barbier-Gauchard (University of Strasbourg); Theophilos Papadimitriou (Democritus University of Thrace)
    Abstract: Since the reinforcement of the Stability and Growth Pact (1996), the European Commission closely monitors public finance in the EU members. A failure to comply with the 3% limit rule on the public deficit by a country triggers an audit. In this paper, we present a Machine Learning based forecasting model for the compliance with the 3% limit rule. To do so, we use data spanning the period from 2006 to 2018 (a turbulent period including the Global Financial Crisis and the Sovereign Debt Crisis) for the 28 EU member states. A set of eight features are identified as predictors from 138 variables through a feature selection procedure. The forecasting is performed using the Support Vector Machines (SVM). The proposed model reached 91.7% forecasting accuracy and outperformed the Logit model that was used as benchmark.
    Keywords: Fiscal Rules, Fiscal Compliance, Stability and Growth Pact, Machine learning.
    JEL: F
    Date: 2022
  2. By: Marc S. Paolella (University of Zurich - Department of Banking and Finance; Swiss Finance Institute); Pawel Polak (Stony Brook University-Department of Applied Mathematics and Statistics)
    Abstract: The CCC-GARCH model, and its dynamic correlation extensions, form the most important model class for multivariate asset returns. For multivariate density and portfolio risk forecasting, a drawback of these models is the underlying assumption of Gaussianity. This paper considers the so-called COMFORT model class, which is the CCC-GARCH model but endowed with multivariate generalized hyperbolic innovations. The novelty of the model is that parameter estimation is conducted by joint maximum likelihood, of all model parameters, using an EM algorithm, and so is feasible for hundreds of assets. This paper demonstrates that (i) the new model is blatantly superior to its Gaussian counterpart in terms of forecasting ability, and (ii) also outperforms ad-hoc three step procedures common in the literature to augment the CCC and DCC models with a fat-tailed distribution. An extensive empirical study confirms the COMFORT model’s superiority in terms of multivariate density and Value-at-Risk forecasting.
    Keywords: GJR-GARCH, Multivariate Generalized Hyperbolic Distribution, Non-Ellipticity, Value-at-Risk.
    JEL: C51 C53 G11 G17
    Date: 2022–11
  3. By: Michał Lewandowski (Faculty of Economic Sciences, University of Warsaw); Marcin Chlebus (Faculty of Economic Sciences, University of Warsaw)
    Abstract: High-quality football predictive models can be very useful and profitable. Therefore, in this research, we undertook to construct machine learning models to predict football outcomes in games from Spanish LaLiga and then we compared them with historical forecasts extracted from bookmakers, which knowledge is commonly considered to be deep and high-quality. The aim of the paper was to design models with the highest possible predictive performances, get results close to bookmakers or even building better estimators. The work included detailed feature engineering based on previous achievements of this domain and own proposals. A built and selected set of variables was used with four machine learning methods, namely Random Forest, AdaBoost, XGBoost and CatBoost. The algorithms were compared based on: Area Under the Curve (AUC) and Ranked Probability Score (RPS). RPS was used as a benchmark in the comparison of estimated probabilities from trained models and forecasts from bookmakers' odds. For a deeper understanding and explanation of the demonstrated methods, which are considered as black-box approaches, Permutation Feature Importance (PFI) was used to evaluate the impacts of individual variables. Features extracted from bookmakers odds’ occurred the most important in terms of PFI. Furthermore, XGBoost achieved the best results on the validation set (RPS equals 0.1989), which obtained similar predictive power to bookmakers' odds (their RPS between 0.1977 and 0.1984). Results of the trained estimators were promising and this article showed that competition with bookmakers is possible using demonstrated techniques.
    Keywords: predicting football outcomes, machine learning, betting, adaboost, random forest, xgboost, catboost, ranked probability score, auc, permutation feature importance
    JEL: C13 C51 C52 C53 C61 L83 Z29
    Date: 2021
  4. By: Greta Goracci; Davide Ferrari; Simone Giannerini; Francesco ravazzolo
    Abstract: Threshold autoregressive moving-average (TARMA) models are popular in time series analysis due to their ability to parsimoniously describe several complex dynamical features. However, neither theory nor estimation methods are currently available when the data present heavy tails or anomalous observations, which is often the case in applications. In this paper, we provide the first theoretical framework for robust M-estimation for TARMA models and also study its practical relevance. Under mild conditions, we show that the robust estimator for the threshold parameter is super-consistent, while the estimators for autoregressive and moving-average parameters are strongly consistent and asymptotically normal. The Monte Carlo study shows that the M-estimator is superior, in terms of both bias and variance, to the least squares estimator, which can be heavily affected by outliers. The findings suggest that robust M-estimation should be generally preferred to the least squares method. Finally, we apply our methodology to a set of commodity price time series; the robust TARMA fit presents smaller standard errors and leads to superior forecasting accuracy compared to the least squares fit. The results support the hypothesis of a two-regime, asymmetric nonlinearity around zero, characterised by slow expansions and fast contractions.
    Date: 2022–11

This nep-for issue is ©2022 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.