Magazine article Financial Management (UK)

Improving Forecasting: Bijan Tabatabai Explains a Method He Developed with IBM to Eliminate Bias and Reduce Random Variation in Business Forecasts

Magazine article Financial Management (UK)

Improving Forecasting: Bijan Tabatabai Explains a Method He Developed with IBM to Eliminate Bias and Reduce Random Variation in Business Forecasts

Article excerpt

Many surveys have shown that companies are often dissatisfied with the quality of their business forecasts. One reason for this is the firms' inability to measure performance effectively in order to provide actionable information on the sources of likely problems.

[ILLUSTRATION OMITTED]

A major cause of quality problems in business forecasting is bias, which is widely recognised but rarely seen as something to tackle strategically. In most cases improving forecasting simply means tackling bias by arbitrary adjustments or second-guessing, (which can add to the problem rather than solving it, since bias patterns can change without notice); or changing the process altogether (which rarely improves matters, since bias can infect the new process).

Unilever is one of the pioneers in seeking a methodical approach to eradicating bias in business forecasting. Most of the problems it has identified concern forecast performance measurement. They include an inability to:

* State definitively whether forecasts are biased or make objective judgments about whether the situation has improved or deteriorated.

* Spot quickly that a forecast is biased--it is little use establishing that it was biased after the period concerned has ended.

* State definitively that the reported risks attached to forecasts are reliable or the range forecasts are the right quality.

To address these issues, I worked with Unilever in 2005-06 to develop the Forecaster's Toolkit, a set of simple tools based on rigorous statistical methods to help users measure and monitor forecast performance objectively (see panel, opposite).

The success of forecasting can be ascertained by looking at how far forecasts deviate from reality. Statistical techniques for analysing forecast errors include mean absolute percentage error (MAPE) and root mean squared error (RMSE). Such techniques are used mainly in statistical forecasting to estimate typical errors in order to appraise or select a model.

Business forecastinq is a judgmental process, but human judgment is inconsistent and is likely to vary according to the specific problem. It is not enough, therefore, to find out the average size of your errors--you also need to know their sources.

Forecast errors are caused by many factors, but the two main types are as follows:

 
Bias                   Random variation 
 
Systematic variation   Unsystematic 
 
caused by              Caused by 
known forces           unknown forces 
 
New or unanticipated   Always present 
 
Unpredictable          Range is predictable 

More importantly, each category requires a different treatment--and applying a single treatment to all of them can have different effects. Walter Shewhart started detecting and separating causes of variation in the twenties as part of a quality-control system for manufacturing and engineering industries. Shewhart categorised causes of process variation as "assignable cause" and "chance cause". Later, William Edwards Deming extended Shewhart's work to improve quality in all areas of an organisation when he developed total quality management. Bias in human judgment (cognitive bias) has also been studied by psychologists and management experts.

In business forecasting, bias can infect a process in two ways:

1 By inputting inappropriate assumptions, systematically but unconsciously, often caused by:

* Framing bias--if your approach is too narrow you will be unable to judge occurrences of future events and their relevance to the forecast.

* Impact bias--consistently overestimating or underestimating the impact of future events.

* Anchoring bias--relying too heavily on selective past events to build forecasts.

* Confidence bias--overestimating or underestimating an activity's potential.

2 By deliberate acts within the process, often caused by:

* Confirmation bias--the tendency to interpret information in such a way as to support a given belief. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.