Exploring Less Widely Used Time Series Models

In the realm of time series forecasting, some models have gained significant traction due to their performance and simplicity, while others have remained lesser-known. This article delves into these less popular models, addressing the gap between scholarly research and practical application.

The Demise of Complexity: What Scholars Prefer

Scholars lean towards more complex models such as ARIMA (AutoRegressive Integrated Moving Average) for their rigorous statistical foundations and robustness in handling various types of time series data. However, this preference might not always align with the needs of practitioners who require models that are easier to implement and explain.

Practitioners and managers, on the other hand, often rely on simpler and more effective methods like Exponential Smoothing or Moving Average. These models not only perform well in forecasting but are also more straightforward, making them user-friendly and cost-effective.

Managers and MBAs: The Art of Forecast Adjustment

While managers and MBA students have the technical capability to use more sophisticated models, they often opt for simplicity due to its practical benefits. However, a common issue arises when they attempt to “touch up” forecasts based on personal intuition or gut feeling.

Quoting a well-known frustration, managers may request analysts to tweak the forecast for specific periods. For example, a manager might say, “Hey, I’ve reviewed your sales forecast and I think October was a bit low; can you adjust it?”

This practice has been studied and proven to be detrimental. Not only does it introduce bias into the forecasting process, but it also undermines the credibility and reliability of the analyst's work. Interactive or iterative adjustments often lead to internal inconsistencies and require the analyst to constantly justify their decisions, which can be both time-consuming and frustrating.

The Role of Exploratory Modeling

In contrast, managers who are open to more complex models are well served by exploring combinations of numerous forecasting techniques. The M Series Competitions, spearheaded by Prof. Spyros Makridakis, a renowned forecaster and statistician, demonstrated the effectiveness of ensemble methods. These competitions involved a mashup of hundreds of models, proving that combining multiple approaches often yields superior results.

The lesson here is to balance the use of advanced techniques with practical considerations. Employing parsimony (simplifying models whenever possible) and understanding the underlying assumptions are key. Extraneous variables can significantly increase costs and complexity, making simpler models more appealing. However, for critical applications, integrating multiple models and focusing on out-of-sample performance metrics such as the PRESS (Predicted Residual Error Sum of Squares) statistic is advisable.

Summary and Practical Advice

To summarize, while scholars may prefer complex models for their theoretic robustness, practitioners often benefit from simpler, more user-friendly methods. Managers and MBAs should be cautious of adjusting forecasts based on personal intuition, as this can lead to unreliable and inconsistent results. On the other hand, exploring ensemble methods and leveraging the wisdom of multiple models can provide significant benefits in terms of accuracy and reliability.

By following these principles, one can strike a balance between theoretical rigor and practical applicability, ensuring that forecasts are both effective and explainable.

Great question!