# How does a forecast bias affect your demand forecast? There are multiple forecasting methods that can be applied in supply chain, namely, linear regression forecasting, exponential smoothing forecasting, etc. Nonetheless, how can we ensure the quality of any forecast? In fact, the quality of any forecast depends on the quality of decisions based on the forecast. So often these decisions are influenced by a bias which results in forecasts that are consistently high or consistently low. Demand forecast shouldn’t be too optimistic or pessimistic, it should be reliable. So, let’s get to know this term by starting with…

#1: What is Forecast bias?

Forecast bias describes a tendency of over-forecasting or under-forecasting. Five key reasons causing forecast bias (Singh, 2021) are the following:

• Sandbagging bias: when there is a system of bonuses based on exceeding of the forecasts
• Anecdote bias: when personal experiences or beliefs are used to support an argument in determining forecasts instead of making decisions based on scientific facts.
• Recent data bias: when an overreaction is created based on the latest events
• Silly bias: when a forecast can be influenced by nonsense objects before you start performing the forecast

#2: How to calculate Forecast bias?

According to Jacobs et al. (2011), forecast bias is measured by a mean error, which is equal to total forecast errors divided by the number of periods where a forecast error per period is equal to the actual demand for the period minus forecasted demand for the same period. Figure 2 illustrates the full formula and an example.

Figure 2. Calculation of a forecast bias (made by the author)

As you can see from the example, there is a chance when forecast errors tend to cancel one another (when error in period 1 = -200 and error in period 2 = 200). This indicates that the measure of forecast bias is likely to be low. The example shows a forecast bias, meaning that the demand was over forecasted by an average of 33 units per period within three periods.

Another formula to measure forecast bias is called Normalized Forecast Metric (Singh, 2021), which is equal to actual demand subtracted by forecast, divided by the sum of forecast and actual demand. The result of this method should be between -1 (under-forecast) and 1, (over-forecast), with 0 indicating of no bias. Over a 12-period horizon, the forecast is biased towards over-forecast when the accumulated values of the metric are more than 2. Likewise, if the accumulated values are less than -2, your forecast tend to be biased towards under-forecast.

#3: How to use Forecast bias to improve forecast accuracy?

There are a few ways to do so. The UK Department of Transportation has developed cost uplifts to forecast costs of different projects. Cost uplift is an increase over the initial estimate. Individual projects receive different cost uplift percentages based on the historical underestimation of their project type. An interesting point is that the forecast is adjusted by someone else rather than the forecaster or demand planner, in order to prevent forecast bias.

Another method is described by Daniel Kahneman in his book “Thinking, Fast and Slow” (2011). The forecast bias prevention process consists of four steps, which are:

1. Determine baseline demand using statistical forecasting techniques: Baseline forecast will help mitigate human impact on the forecast
2. Create an intuitive forecast based on available evidence/ current events (e.g. new market, discount)
3. Correlation estimate: Identify to what extent the factors used to create the intuitive forecast capture all the factors that drive demand. In many cases, intuitive forecast doesn’t cover the whole story. There are other internal and external factors, such as competitor’s actions, natural disaster, production shutdown, etc.
4. Adjust the baseline forecast based on the estimation: Let’s say your baseline forecast is 100, your intuitive forecast is 200, and the correlation estimate is 25% (your intuitive forecast accounts for all factors driving demand. This means the adjusted forecast should be 150 units.

John Parks, IBM Senior Managing Consultant, suggested identifying the pattern in error and performing root cause analysis either quantitatively or qualitatively via collaboration. James Bentzley provided a similar method, in which organizations should examine the aggregate forecast and then drill deeper to products at the lowest levels. He also indicated to investigate a likelihood of negative/positive bias when demand was greater/ lower than forecast for three or more months in a row. The bias is removed when actual demand bounces back and forth with regularity above and below forecast.

Thank you very much for reading and please share if you have experienced with forecast bias and how you dealt with it!

To sum up, I believe the following quote will give you an impression of the importance of forecast bias:

“Kakouros, Kuettner and Cargille provide a case study of the impact of forecast bias on a product line produced by HP. They stated that eliminating bias from forecasts resulted in a 20 to 30% reduction in inventory while still maintaining high levels of product availability. Similar results can be extended to the consumer goods industry where forecast bias is prevalent”. – APICS (2002)