Enhancing Predictive Accuracy in Equity Markets: An Empirical Analysis of Outlier Robustness and Forecasting Efficiency in Dow Jones Index Returns
https://doi.org/10.55966/assaj.2025.4.1.065
Abstract
This paper empirically examines methods for the detection of outliers and their effect on forecast accuracy in financial time series data using a number of criteria that have been found to discriminate among forecast methods. We have used daily returns from the Dow Jones Index (April 2002–June 2024) where we divide the sample into an in-sample estimation period (April 2002–May 2023) and an out-of-sample (1-year) forecasting window (June 2023–June 2024). In this paper, we also compare the above six outlier detection methods with three volatility forecasting models (GARCH and its extensions) to measure the impact of those synergistic effects on prediction.
We show that outlier removal systematically improves the precision of models, reflected in reductions in average mean absolute error (MAE) and mean absolute percentage error (MAPE). The MADe Method is found to be the best outlier detection method, and the three models produce better predictions in all cases (GARCH: ΔMAPE = −18.3%, ΔMAE = −12.7%; EGARCH: ΔMAPE = −14.9%; GJR: ΔMAPE = −11.2%) also in both fields two auxiliary methods, namely Tukey’s Method and VH Boxplot, are respectively the second and third best outlier detection method. We observe that, in both outlier-adjusted and unadjusted cases, their GARCH versions consistently exhibit better performance compared to the respective asymmetric models (i.e., EGARCH and GJR), suggesting a better ability of GARCH under standard setting to address leptokurtic shocks but when associated with MADe-based filtration.
The best process, which includes MADe outlier method within GARCH(1, 1) forecasting, decreases overall autocorrelation forecast errors by 22.4% below benchmark levels. The obtained findings have important implications for quantitative finance for the purpose of developing better approaches to modeling volatility and for improving risk management in equity markets.
Keywords: Equity Markets, Outliers Detection, MADe Method, Forecasting, GARCH Models.
