[Paper Review] Forecasting Economics and Financial Time Series: ARIMA vs. LSTM

Sima Siami‐Namini, Akbar Siami Namin|arXiv (Cornell University)|Mar 16, 2018

Stock Market Forecasting MethodsDecision Sciences22 references59 citations

TL;DR

This paper compares ARIMA with LSTM for forecasting economics and financial time series and finds that LSTM substantially outperforms ARIMA in predictive accuracy, with an average error reduction of 84–87%. It also reports that training epochs have no consistent effect on performance.

ABSTRACT

Forecasting time series data is an important subject in economics, business, and finance. Traditionally, there are several techniques to effectively forecast the next lag of time series data such as univariate Autoregressive (AR), univariate Moving Average (MA), Simple Exponential Smoothing (SES), and more notably Autoregressive Integrated Moving Average (ARIMA) with its many variations. In particular, ARIMA model has demonstrated its outperformance in precision and accuracy of predicting the next lags of time series. With the recent advancement in computational power of computers and more importantly developing more advanced machine learning algorithms and approaches such as deep learning, new algorithms are developed to forecast time series data. The research question investigated in this article is that whether and how the newly developed deep learning-based algorithms for forecasting time series data, such as "Long Short-Term Memory (LSTM)", are superior to the traditional algorithms. The empirical studies conducted and reported in this article show that deep learning-based algorithms such as LSTM outperform traditional-based algorithms such as ARIMA model. More specifically, the average reduction in error rates obtained by LSTM is between 84 - 87 percent when compared to ARIMA indicating the superiority of LSTM to ARIMA. Furthermore, it was noticed that the number of training times, known as "epoch" in deep learning, has no effect on the performance of the trained forecast model and it exhibits a truly random behavior.

Motivation & Objective

• Assess the relative forecasting accuracy of traditional time series models (ARIMA) versus deep learning approaches (LSTM) for economics and finance data.
• Quantify the average improvement in prediction error when using LSTM over ARIMA.
• Investigate how training dynamics (epochs) affect the performance of LSTM models in time series forecasting.

Proposed method

• Review and empirical comparison of ARIMA and LSTM forecasting on economics/finance time series.
• Use standard forecasting error metrics to quantify performance differences.
• Analyze the impact of training epochs on LSTM performance (finding that epoch count shows random behavior/no effect).

Experimental results

Research questions

• RQ1 Do LSTM models outperform ARIMA in forecasting economics and financial time series?
• RQ2 What is the magnitude of error reduction when using LSTM compared to ARIMA?
• RQ3 How do training epochs influence the forecasting performance of LSTM in this domain?

Key findings

• LSTM outperforms ARIMA in forecasting economics and financial time series.
• Average reduction in error rates when using LSTM is reported as 84–87 percent versus ARIMA.
• The number of training epochs shows no consistent effect on LSTM performance (random behavior).