nep-for New Economics Papers
on Forecasting
Issue of 2023‒12‒04
four papers chosen by
Rob J Hyndman, Monash University

  1. ARIMA and LSTM: A Comparative Analysis of Financial Time Series Forecasting By Joao Vitor Matos Goncalves; Michel Alexandre; Gilberto Tadeu Lima
  2. Solving the Forecast Combination Puzzle By David T. Frazier; Ryan Covey; Gael M. Martin; Donald S. Poskitt
  3. Trading on short-term path forecasts of intraday electricity prices. Part II -- Distributional Deep Neural Networks By Grzegorz Marcjasz; Tomasz Serafin; Rafal Weron
  4. Optimal Forecast Combination with Mean Absolute Error Loss By Felix Chan; Laurent Pauwels

  1. By: Joao Vitor Matos Goncalves; Michel Alexandre; Gilberto Tadeu Lima
    Abstract: This paper assesses the impact of time horizon on the relative performance of traditional econometric models and machine learning models in forecasting stock market prices. We employ an extensive daily series of Brazil IBX50 closing prices between 2012 and 2022 to compare the performance of two forecasting models: ARIMA (autoregressive integrated moving average) and LSTM (long short-term memory) models. Our results suggest that the ARIMA model predicts better data points that are closer to the training data, as it loses predictive power as the forecast window increases. We also find that the LSTM model is a more reliable source of prediction when dealing with longer forecast windows, yielding good results in all the windows tested in this paper.
    Keywords: Finance; machine learning; deep learning; stock market
    JEL: C22 C45 C53 G17
    Date: 2023–11–17
  2. By: David T. Frazier; Ryan Covey; Gael M. Martin; Donald S. Poskitt
    Abstract: The forecast combination puzzle is the commonly encountered empirical result whereby predictions formed by combining multiple forecasts in complex ways do not out-perform more naive, e.g. equally-weighted, approaches. While various solutions for the cause of the puzzle exist in the literature, these solutions are limited in their scope and applicability. In contrast, we demonstrate a general solution to the puzzle by showing that this phenomenon is a direct consequence of the methodology used to produce forecast combinations. In particular, we show that tests which aim to discriminate between the predictive accuracy of competing forecast combination strategies have low power, and can lack size control, leading to an outcome that favours the naive approach. In addition, we demonstrate that the low power of such predictive accuracy tests in the forecast combination setting can be completely avoided if more efficient strategies are used in the production of the combinations. We illustrate these findings both in the context of forecasting a functional of interest and in terms of predictive densities. A short empirical example using daily financial returns exemplifies how researchers can avoid the puzzle in practical settings.
    Keywords: optimal forecast combinations, tests for forecast accuracy, probabilistic forecasting, scoring rules, S&P500 forecasting, one-step versus two-step estimation
    JEL: C18 C12 C53
    Date: 2023
  3. By: Grzegorz Marcjasz; Tomasz Serafin; Rafal Weron
    Abstract: We propose a novel electricity price forecasting model tailored to intraday markets with continuous trading. It is based on distributional deep neural networks with Johnson SU distributed outputs. To demonstrate its usefulness, we introduce a realistic trading strategy for the economic evaluation of ensemble forecasts. Our approach takes into account forecast errors in wind generation for four German TSOs and uses the intraday market to resolve imbalances remaining after day-ahead bidding. We argue that the economic evaluation is crucial and provide evidence that the better performing methods in terms of statistical error metrics do not necessarily lead to higher trading profits.
    Keywords: Intraday electricity market; Probabilistic forecast; Path forecast; Prediction bands; Trading strategy; Neural networks
    JEL: C22 C32 C45 C51 C53 Q41 Q47
    Date: 2023
  4. By: Felix Chan; Laurent Pauwels
    Abstract: The optimal aggregation of forecasts produced either from models or expert judgements presents an interesting challenge for managerial decisions. Mean absolute error (MAE) and mean squared error (MSE) losses are commonly employed as criteria of optimality to obtain the weights that combine multiple forecasts. While much is known about MSE in the context of forecast combination, less attention has been given to MAE. This paper shows that the optimal solutions from minimizing either MAE or MSE loss functions, i.e., the optimal weights, are equivalent provided that the weights sum to one. The equivalence holds under mild assumptions and includes a wide class of symmetric and asymmetric error distributions. The theoretical results are supported by a numerical study that features skewed and fat-tailed distributions. The practical implications of combining forecasts with MAE and MSE optimal weights are investigated empirically with a small sample of data on expert forecasts on inflation, growth, and unemployment rates for the European Union. The results show that MAE weights are less sensitive to outliers, and MSE and MAE weights can be close to equivalent even when the sample is small.
    Keywords: forecasting, forecast combination, optimization, mean absolute error, optimal weights
    JEL: C53 C61
    Date: 2023–11

This nep-for issue is ©2023 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.