Invariably inaccurate: Why is forecasting still matter?

Are forecasts consistently wrong?

It is undeniable that people are unable to predict the future with 100% accuracy, even with the support of Tarot cards. Can anyone forecast how Covid-19 has caused things to deviate from their original trajectory?. This shows that forecasting is an extremely difficult task. 

Many reasons are considered when it comes to explaining why it is difficult to achieve absolutely accurate forecasts, some can be listed below:

• Inappropriate models/methods: The adopted model may perfectly match past data but unnecessarily match the future. The economic crisis is a concrete instance. Before the crisis, an incredible upward trend in the market was witnessed. In that case, if the same old model with the same trend line was repeatedly applied, the forecasted economic outlook would be fairly positive contrasting to the non-stop plunge in reality.  This partly confirms that choosing the right forecasting model is imperative because completely reversed results would be shown with an unsuitable one. 

  

• The level of the forecaster: According to the aforementioned analysis, it can be concluded that forecasting is relatively difficult. Despite the help of state-of-the-art computer systems, effective operation still requires humans. Therefore, if the forecasters are not well-qualified as well as do not have the necessary techniques and flexibility, the forecast results will be greatly affected.
• High expectations on the forecast accuracy: According to the theory of statistical probability, high expectations will lead to a greatly stretched forecast range. It means that, when you want to achieve 100% accuracy, the results obtained for the forecast value will be in the range of 0 to infinity, which is much wider than that of 95% or less accuracy expectation. Obviously, too high expectations make the process as well as the predicted value reduce its meaning.

Although the aforementioned causes can be overcome after many adjustments, the deviation caused by the impact of random factors is extremely difficult to control. Unfortunately, they always appear in most forecast results. Therefore, it is stated that forecasts are never accurate since there are always random errors. Nowadays, forecasting becomes more and more difficult owing to the constantly changing market which is counted by hours, minutes or even seconds. As a consequence, the inaccurate forecast could be even more inaccurate. 

What should forecasting be approached?

It is certain that rarely (and almost never)do forecasts coincide with reality. Accepting the fact that the forecast results are 100% accurate is unimaginable, but for forecasters, results with 95% accuracy, whose error values are within the allowable range have already been amazing. The always wrong forecast results do not coincide with worthless values, the thing is the  forecasters’ capability of controlling the deviation, which tackles the issues. 

“Thinking forecast accuracy as the inverse of forecast error is a major problem”

For instance, a business forecasts demand for item A for the following month and gets the result of 1000 more units needed to be produced in order to satisfy all customer demand. Additionally, managers also take the forecast accuracy (95%) (equivalent to 5% error) into account. As a result, different strategies will be adopted by businesses, but  5%x1000=50 more product units could be regarded as the most basic level of inventory held in the company’s warehouse with a view to preventing shortages. However, this is just a possibility, in fact, the error can be consulted for a flexible approach, as long as the solution fulfills the business’s preset goals (meet all customers’ requirement, in this case)

The measures of forecast errors

In order to analyse and track the forecast improvement over time, a customized metrics system should be developed. Several metrics are utilized to calculate forecast accuracy, most of which are based on the following three metrics: forecast bias, mean absolute deviation (MAD) and mean absolute percentage error (MAPE).

• Forecast bias

The forecast bias is the difference between the forecast result and the actual value. Forecast results are considered as positive if they exceed the estimated sales level, and vice versa. Comparing the ratio of total forecast to total sales with 100% is a way to inspect the bias. If the bias is higher than 100%, the forecast is overrated, in contrast,  it is underrated.

Nevertheless, deviations for long-term forecasting or multiple-product forecasting do not provide detailed information, therefore, this method is normally applied to aggregated forecasts rather than the disaggregated ones.

• Mean Absolute Deviation (MAD)

The mean absolute deviation (MAD) is the average of the absolute forecast errors over time. The MAD indicates the average deviation of the forecast, the smaller this value, the higher the accuracy. This figure which provides an assessment of the entire forecasting process instead of only a few recent results will support  managers in making more optimal decisions. 

• Mean Absolute Percentage Error (MAPE)

Basically, MAPE shows the average percentage deviation of forecast compared to reality. MAPE is a good measure when the forecasted value is affected by seasonal and periodical factors. In several cases, the two above-mentioned indicators do not fully represent the forecast errors’ multi-perspectives since only absolute deviation is shown, MAPE is the solution, which relates the difference to the actual factor in order to equip managers with the most objective view on the forecast result.

Which indicator is the best?

It is such a challenging question since each forecaster has their own answer depending on the forecast purpose, aggregation level and the forecast period, each indicator will provide different information.

Take an example, the forecast level in the retail store level forecast is completely different from that in the distribution center. With regards to the store, the forecasting requirement is for each detailed product. While in the distribution center, the process focuses on large product lines and more composite elements. Hence, it can be concluded that different forecasts will be applied for different purposes and objects, which leads to a vary in accuracy indicators.

The importance of forecasting

Now let’s go back to the original question “Why are forecasts always wrong, but companies still invest in this?”

Paul Staffo once said that the purpose of forecasting is not to predict the future, but to help managers take action that is useful in the present. Forecasting is the basis for companies to quickly respond to market fluctuations. A good forecast ensures manufacturers maintain sufficient raw material input for effective and efficient operations. In fact, a few technique adjustments to the results are required before any decisions are made. Due to the numerical methods’ incapability of predicting all random factors, the results must be adjusted by well-qualified and experienced managers to accomplish expected outcome. Therefore, there is the statement: “Forecasting is both science and art”.

“The goal of forecasting is not to predict the future but to tell you what you need to know to take meaningful action in the present” – Paul Staffo

From the above analysis, it can be concluded that forecasting is regarded as a guideline for business activities. Once acknowledging the characteristics of forecasts which indicate that forecasting is invariably inaccurate, managers are well-prepared to confront unpredictable events. From a different viewpoint, only a certain percentage of accuracy is still much better than nothing. Despite the inaccuracy, forecasting’s positive impact is indisputable. As a consequence, forecasting is still utilized by businesses although the results are only correct on paper. 

Huyen Tran, Nhat Huyen