In week 6 of the course we will look at demand management and forecasting, an area that is receiving substantial attention, especially as interest in supply chain management grows and we seek to more effectively plan and coordinate the supply chain as a whole.
It is often said that forecasts are usually wrong, some spectacularly so:
The learning objectives for this week of the course are that you should understand the role of forecasting as a basis for supply chain planning. That you will be able to compare the differences between independent and dependent demand. Thirdly, that you will be able to identify the basic components of independent demand, including average, trend, seasonal and random variation. You will be able to describe the common qualitative forecasting techniques such as Delphi Method and Collaborative Forecasting. You will understand basic quantitative forecasting techniques and the use of decomposition to forecast when trend and seasonality is present.
The following video emphasises the need for accuracy and commonsense in forecasting:
Forecasts can be divided into two types, strategic and tactical. Strategic forecasts are used to assist the creation of the strategy that will determine how demand is met. Tactical forecasts are used to assist decision making on a day to day basis. Demand management is used to influence the sources of product or service demand, either increasing demand, decreasing demand or maintaining it at a constant level. The following video looks at the factors that influence forecasting in the wine industry:
Dependent and Independent Demand
There are two basic sources of demand, dependent and independent. Dependent demand is the demand that occurs as a result of the demand for other products or services. Independent demand is demand that cannot be forecast based on the demand for another product or service.
Dependent demand is usually very difficult to influence – it is demand that is not dependent on factors that you can influence and rather it is demand that you have to meet. Independent demand can usually be influenced and therefore organisations have a choice about whether they take an active role and influence it or take a passive role and simply respond to the demand that exists. The following video looks at how Motorola work with their forecasting:
The textbook identifies four basic types of forecasts. Qualitative forecasting is based on human judgement and some of the techniques used in qualitative forecasting will be discussed below. Time series analysis looks at patterns of data over time. Causal relationships looks at the relationships between factors that will influence demand and simulation seeks to model demand so that the inter-relationship of demand factors can be better understood. The following video examines how demand management and forecasting are undertaken at Lowes:
Usually demand is thought of as having six components, average, trend, seasonal elements, cyclical elements, random variation and autocorrelation. These elements of demand enable us to understand the pattern of demand for a product that might be applied to the prediction of future demand.
Average demand is the average demand for a product over time. The trend shows how demand has changed over time and seasonal demand shows seasonal variations in demand. Cyclical elements occur over a longer period than seasonal elements and are harder to predict, occurring, for example, as a result of economic cycles. Random variation is based on random events that are impossible to predict while auto-correlation is the relationship between past and future demand, that is, that future demand is related to current demand. Where there is a high degree of random variation there is very little relationship between current demand and future demand. Where there is a high degree of auto-correlation there is a strong relationship between current and future demand.
Time Series Models
Time series models forecast the future based on past models. Various models are available and the one that you should use depends on the time horizon that you wish to forecast, the data that you have available, the accuracy that you require, the size of the forecasting budget and the availability of suitably qualified people to undertake the analysis. The following chart from page 488 of the textbook is design to assist with selecting the appropriate tool:
Linear regression is used where there is a functional relationship between two correlated variable, being used to predict one variable based on the other. It is useful where data is relatively stable.
Decomposition of a time series is used to identify and separate the time series data into its various demand components. Two types of seasonal variation are identified – additive, where the seasonal amount in each season is constant and multiplicative where the seasonal variation is a percentage of the demand for a time period.
The simple moving average is useful when demand is relatively stable, not increasing or decreasing rapidly and where there are few seasonal characteristics. Moving averages can be centred around their midpoint, or used as a basis for predicting the future. Using a longer time period will result in more smoothing of variation while using a shorter time period will reveal statistical trends more quickly.
A weighted moving average allows you to weight particular time periods within the average to achieve greater accuracy. For example, heavier weight may be given to more recent time periods in order to place more emphasis on recent demand activity.
Exponential smoothing is the most used of all forecasting techniques and appears in all computer based forecasting applications. It is used alot in retail and service industries. It is often very accurate, it is fairly easy to do, it is easily understood, requires little computation and is easily tested for accuracy.
The following video details the conduct of these forecasting techniques:
Qualitative forecasting involves applying human judgement to create a forecast. Usually a structured approach is used, unlike this:
Various techniques are used for qualitative forecasting, including:
Historical Analogy: Basing forecasts on the demand pattern for similar products.
Market Research: Forecasts are created by a market research company, mainly using surveys and interviews.
Panel Consensus: Where a group of people with knowledge in the forecast subject area, share their thoughts and develop a forecast.
Delphi Method: A survey based technique that creates anonymity in a group. It is described in the following video:
Collaborative Planning, Forecasting and Replenishment: CPFR is a recent innovation that uses the internet to allow people to collaborate on forecast creation:
There are two types of forecast errors. Bias errors occur where there is a consistent mistake made that permeates the forecast made. Random errors are errors that can’t be explained by the forecast model – they occur randomly and on an unpredictable basis. Measures of forecast error include Mean Absolute Deviation (MAD), Mean Absolute Percentage Error (MAPE) and Tracking Signal. The following video considers issues in human forecast error:
Tracking Signal is a measure that is used to monitor the actual performance of the forecast over time to see whether it is in line with the changes in demand in the real world. It can be used like a quality control chart.
This week we have considered demand management and forecasting, using both qualitative and quantitative techniques. Emphasis has been put on ensuring that forecasts are realistic and caution has been advised on the use of forecasting based on past performance – it doesn’t usually tell you what the future will do but often will help you to prepare. The following video features the application of information technology to forecasting and is perhaps a humorous conclusion to this week’s material: