What are the Best Demand Forecasting Methods for Uncertain Situations?

The core of any demand planning process is a statistical forecast. These models act as the fundamental building blocks in the planning process, guiding decision-making for businesses across various industries. Statistical forecasting is the initial step in establishing a baseline plan, which can then be refined through consensus and collaboration. Ensuring a thorough process and appropriate software technologies are crucial for accuracy.

Types of Forecasting Techniques

Forecasting techniques can be broadly categorized into several groups, each suited for different scenarios and data types. Here's a breakdown of the main methods:

Time Series Models

Time series models rely on historical data to predict future values. Common examples include:

Exponential Smoothing Holt-Winters Multiplicative Smoothing ARIMA (AutoRegressive Integrated Moving Average) Models Box-Jenkins Models Logarithmic Regression Models

These models are particularly useful when dealing with stationary or nearly stationary time series data.

Promotional Planning Models

Promotional planning models typically incorporate event modeling methodologies, where specific events such as marketing campaigns or holidays are accounted for. Indicator variable models are also used to represent these events. These models are essential for businesses that rely heavily on promotional strategies to drive demand.

Causal Models

Causal models examine the relationship between the demand and various explanatory variables. Multiple Linear Regression Models and Transfer Function Models fall into this category. These models are ideal when there are several factors that influence demand, such as price, seasonality, and external economic indicators.

Probabilistic Models

Probabilistic models focus on forecasting the probability of events occurring in the future. Techniques like Logit, Probit, and Tobit models are used to predict the likelihood of specific outcomes. These models are particularly useful in scenarios where demand variability is high and uncertain.

Croston's Method

Croston's Method is specifically designed for intermittent demand forecasting, where demand is sporadic and unpredictable. This method divides demand into two components: demand size and demand interval, and forecasts these separately. For a detailed understanding, you may refer to the semi-technical explanation provided in the additional resource.

Factors to Consider

When choosing the best forecasting method, several factors should be taken into account:

Non-Data Based Information: Any additional knowledge or insights that might influence the forecast. Seasonality: The presence or absence of seasonal effects in the data. Data Stability: Whether the data is stationary or can be made stationary using techniques like ARMA, ARIMA, SARIMA, or ARIMAX. External Variables: Other variables that can help in explaining the demand series, which may include marketing efforts, economic indicators, or changes in market conditions. Data Set Size: The size and quality of the data set can significantly impact the accuracy of forecasts.

For a comprehensive guide to demand forecasting, you might want to read the book Forecasting: Principles and Practice by Rob J Hyndman and George Athanasopoulos. This book, available for free on the web, offers a wealth of knowledge on various forecasting methods and their practical applications.

Evaluating Forecast Models

When evaluating different forecast models, it is essential to use techniques that go beyond simple in-sample fit. A common method involves splitting the data into training and testing sets, as follows:

For a one-period ahead forecast, use the first m - n observations to estimate the candidate model and forecast the m - n 1 observation. n - m estimated errors. Use these estimated errors to evaluate the forecasting model.

Many forecasting software packages include routines for completing such processes. However, it's important to consult your software manual for variations and further guidance.

Combining Statistical and Non-Statistical Forecasts

While statistical models provide valuable insights, it's crucial not to ignore the use of non-statistical, judgemental forecasts. Judgemental models rely on the insights and experience of experts, which can be particularly useful when dealing with complex or unique scenarios.

For a detailed discussion on the integration of judgemental forecasts, you can refer to the chapter on this topic in the same reference book mentioned earlier.

By considering these factors and methods, you can better prepare for uncertain situations and make more accurate demand forecasts.