Abstract:

The simple moving average (SMA) is a cornerstone of time series forecasting, prized for its simplicity and interpretability. However, its standard formulation is notoriously inadequate for data characterized by significant seasonality and trends, often resulting in lagged and inaccurate forecasts. This research addresses this critical limitation by developing and evaluating a suite of advanced moving average-based models specifically designed to decompose and capture these complex components. We propose a hybrid framework that integrates classical decomposition techniques with adaptive moving average filters. The methodology involves: (1) applying seasonal differencing or seasonal adjustment to isolate the trend-cycle, (2) utilizing double or triple moving averages to project the underlying trend, and (3) incorporating seasonal indices to reinstate periodic fluctuations. The performance of these advanced models—including Holt-Winters-inspired moving average adaptations—is rigorously tested against the standard SMA and more complex benchmarks like SARIMA and ETS models. Using a diverse set of synthetic and real-world datasets with known seasonal and trend patterns, our empirical analysis demonstrates that the proposed advanced moving average models achieve a substantial reduction in forecast error compared to the simple moving average. While not universally superior to sophisticated statistical models, they offer a compelling trade-off, providing a significant boost in accuracy with only a marginal increase in computational complexity. The findings indicate that these enhanced techniques are a highly viable and accessible forecasting tool for practitioners in business and economics, bridging the gap between simplistic and statistically complex methods.

Description:

.