nep-for New Economics Papers
on Forecasting
Issue of 2023‒10‒23
three papers chosen by
Rob J Hyndman, Monash University

  1. Univariate Forecasting for REIT with Deep Learning: A Comparative Analysis with an ARIMA Model By Axelsson, Birger; Song, Han-Suck
  2. Short-Term Stock Price Forecasting using exogenous variables and Machine Learning Algorithms By Albert Wong; Steven Whang; Emilio Sagre; Niha Sachin; Gustavo Dutra; Yew-Wei Lim; Gaetan Hains; Youry Khmelevsky; Frank Zhang
  3. Transformers versus LSTMs for electronic trading By Paul Bilokon; Yitao Qiu

  1. By: Axelsson, Birger (Department of Real Estate and Construction Management, Royal Institute of Technology); Song, Han-Suck (Department of Real Estate and Construction Management, Royal Institute of Technology)
    Abstract: This study aims to investigate whether the newly developed deep learning-based algorithms, specifically Long-Short Term Memory (LSTM), outperform traditional algorithms in forecasting Real Estate Investment Trust (REIT) returns. The empirical analysis conducted in this research compares the forecasting performance of LSTM and Autoregressive Integrated Moving Average (ARIMA) models using out-of-sample data. The results demonstrate that in general, the LSTM model does not exhibit superior performance over the ARIMA model for forecasting REIT returns. While the LSTM model showed some improvement over the ARIMA model for shorter forecast horizons, it did not demonstrate a significant advantage in the majority of forecast scenarios, including both recursive multi-step forecasts and rolling forecasts. The comparative evaluation reveals that neither the LSTM nor ARIMA model demonstrated satisfactory performance in predicting REIT returns out-of-sample for longer forecast horizons. This outcome aligns with the efficient market hypothesis, suggesting that REIT returns may exhibit a random walk behavior. While this observation does not exclude other potential factors contributing to the models' performance, it supports the notion of the presence of market efficiency in the REIT sector. The error rates obtained by both models were comparable, indicating the absence of a significant advantage for LSTM over ARIMA, as well as the challenges in accurately predicting REIT returns using these approaches. These findings emphasize the need for careful consideration when employing advanced deep learning techniques, such as LSTM, in the context of REIT return forecasting and financial time series. While LSTM has shown promise in various domains, its performance in the context of financial time series forecasting, particularly with a univariate regression approach using daily data, may be influenced by multiple factors. Potential reasons for the observed limitations of our LSTM model, within this specific framework, include the presence of significant noise in the daily data and the suitability of the LSTM model for financial time series compared to other problem domains. However, it is important to acknowledge that there could be additional factors that impact the performance of LSTM models in financial time series forecasting, warranting further investigation and exploration. This research contributes to the understanding of the applicability of deep learning algorithms in the context of REIT return forecasting and encourages further exploration of alternative methodologies for improved forecasting accuracy in this domain.
    Keywords: Forecasting; Equity REITs; deep learning; LSTM; ARIMA
    JEL: G17 G19
    Date: 2023–09–28
  2. By: Albert Wong; Steven Whang; Emilio Sagre; Niha Sachin; Gustavo Dutra; Yew-Wei Lim; Gaetan Hains; Youry Khmelevsky; Frank Zhang
    Abstract: Creating accurate predictions in the stock market has always been a significant challenge in finance. With the rise of machine learning as the next level in the forecasting area, this research paper compares four machine learning models and their accuracy in forecasting three well-known stocks traded in the NYSE in the short term from March 2020 to May 2022. We deploy, develop, and tune XGBoost, Random Forest, Multi-layer Perceptron, and Support Vector Regression models. We report the models that produce the highest accuracies from our evaluation metrics: RMSE, MAPE, MTT, and MPE. Using a training data set of 240 trading days, we find that XGBoost gives the highest accuracy despite running longer (up to 10 seconds). Results from this study may improve by further tuning the individual parameters or introducing more exogenous variables.
    Date: 2023–05
  3. By: Paul Bilokon; Yitao Qiu
    Abstract: With the rapid development of artificial intelligence, long short term memory (LSTM), one kind of recurrent neural network (RNN), has been widely applied in time series prediction. Like RNN, Transformer is designed to handle the sequential data. As Transformer achieved great success in Natural Language Processing (NLP), researchers got interested in Transformer's performance on time series prediction, and plenty of Transformer-based solutions on long time series forecasting have come out recently. However, when it comes to financial time series prediction, LSTM is still a dominant architecture. Therefore, the question this study wants to answer is: whether the Transformer-based model can be applied in financial time series prediction and beat LSTM. To answer this question, various LSTM-based and Transformer-based models are compared on multiple financial prediction tasks based on high-frequency limit order book data. A new LSTM-based model called DLSTM is built and new architecture for the Transformer-based model is designed to adapt for financial prediction. The experiment result reflects that the Transformer-based model only has the limited advantage in absolute price sequence prediction. The LSTM-based models show better and more robust performance on difference sequence prediction, such as price difference and price movement.
    Date: 2023–09

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.