Demand Forecasting

Demand forecasting is an activity a company does internally when it sets its sales budget. The demand forecast influences all upstream commitments and decisions. Forecasting is important and fundamental to any business. It is the act of looking ahead and anticipating the future.

Forecasting provides lead time to do the following:

• Respond to new situations (avoid surprises)
• Make optimal, and proactive decisions, instead of doing things by default, in a reactive mode

Types of demand forecasts

Long-term

Long-term forecasts are for strategic management decisions such as those concerning new product introduction, large investments, acquisitions, entry into new regions or markets, and more.

Medium-term

Medium-term forecasts relate to tactical, yearly, decisions. These include inventory planning, master production planning, subcontracting policies, hiring, setting staff/sales targets and bonuses, and more.

Short-term

Short-term forecasts are for daily and weekly scheduling.

Hazards

Overestimation

An overestimation of demand can result in:

• Excess resources
• Excess inventories
• The need to have discount sales

Underestimation

An underestimation of demand can result in various opportunity costs including:

• Lost sales
• Loss of goodwill
• Damage of company image
• Loss of market share
• Loss of revenue presently and in the future

Methods

Qualitative

The qualitative method of demand forecasting can use various models. They are:

• Management judgment
• Expert opinions
• Salesforce composites
• Delphi method
• More

Delphi method

The Delphi method is an anonymous group forecasting exercise. None of the participants knows who the others are. The exercise is managed by a Delphi coordinator.

By being anonymous, bias is removed from members that would otherwise skew the data. However, the Delphi method may require many sessions before a consensus is reached.

This method is usually employed for new products, technologies, and industry forecasts.

Quantitative

Time series

Time series forecasting uses historical figures to predict future results. For example, a restaurant may use last month’s sales figures to predict how much food it will sell the next month.

One of the drawbacks of time series forecasting is that it assumes the future will be the same (or similar) to the past. It does not address any other variables.

A time series has four major elements:

1. Seasonality. These can be daily, weekly, monthly, or quarterly. For example, most car washes are busiest on the weekends. Or shops that sell ski equipment are busiest in the winter.
2. Trends. A trend is a slight increase or decreases over time. Trends can be affected by economic conditions, fads, demographics, and more.
3. Cycles. These are fluctuations that repeat every few years. They can be due to economics, politics, and the general business environment. Cycles are difficult to forecast.
4. Random variables. These are unpredictable and unexpected sudden increases or decreases in the data points. They have no pattern.

In the quantitative time series method of demand forecasting, the x-axis (horizontal) = time. Various models include:

• Simple and weighted moving average
• Exponential smoothing
• Exp. smoothing adjusted for trends
• Trend-seasonality regression
• Other models

Causal regression

In the quantitative causal regression method of demand forecasting, the independent variables = causal variables. The independent variables are those that the firm can manipulate.

Macro forecasts

There are many large-scale forecasts which can be useful in looking forward to future operations. They include:

• Consumer confidence index
• Gross domestic product (GDP)
• Consumer price index (CPI)
• Housing starts (number of new privately-owned homes)
• Personal income and consumption
• Producer price index (PPI)
• Index of leading economic indicators
• Purchasing manager’s index
• Retail sales

Measuring accuracy

It is important to be able to measure the accuracy of forecasts. If a company’s forecasts have been accurate in past periods, they should remain so. If they have not been accurate, it is worth understanding how they have been wrong and where to correct them.

Simple measures

In general, the accuracy can be determined by subtracting the findings from the actual results:

Error = Actual - Forecast

For example, if a company forecast it would sell $10 million, and it only sold $8 million, we can see that the error is $-2 million.

Or, an absolute error can be determined:

Absolute Error = | Actual - Forecast |

Spread measures

This type of measure shows an absolute figure (not positive or negative).

Ft represents the forecast in period t.

At represents the demand in period t.

Sigma means the sum of.

MAD

Mean absolute deviation (MAD) is a measure of a model’s forecast error. It shows how accurate the model is. It is the sum of the absolute values of each forecast error, divided by the number of time periods (n).

Mean Absolute Deviation (MAD) = Mean Absolute Error
= (Sigma| At - Ft|)/n 
MAD is around 0.8 times std. deviation

MAPD

Mean Absolute % Deviation (MAPD) 
= ( MAD / Avg. Sales) * 100%

MSE

Mean squared error is another way of calculating the overall accuracy of a forecast model. It is the average of the squared differences of the forecast and actual values.

The formula is as follows:

Mean Squared Error (MSE) =   Sigma(At - Ft)squared/n

Because of the squaring involved, MSE can often make large errors show up to a significant degree.

Bias measure

This type of measure shows the positive or negative figure (keeps the sign of error).

A tracking signal is the rolling sum of forecast errors / MAD. The ideal value is zero. A value of zero means that the demand forecasting is right on target.

A high positive tracking signal = consistent underestimation

A high negative tracking signal = consistent overestimation

Cite Term

Definition Sources