Scribd
Upload
36 views7 pages

SCM - Demand Forecasting

The document discusses demand forecasting methods including judgemental, experimental, causal, and projective methods. Common projective methods are moving average and exponential smoothing. Advanced methods include double/triple exponential smoothing and neural networks. The document outlines steps to take a methodical approach to demand forecasting including planning, data checking, product categorization, using metrics, and controlling forecasts.

Uploaded by

GRS
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
36 views7 pages

SCM - Demand Forecasting

The document discusses demand forecasting methods including judgemental, experimental, causal, and projective methods. Common projective methods are moving average and exponential smoothing. Advanced methods include double/triple exponential smoothing and neural networks. The document outlines steps to take a methodical approach to demand forecasting including planning, data checking, product categorization, using metrics, and controlling forecasts.

Uploaded by

GRS
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Demand Forecasting

Definition
Demand forecasting is a set of methods used to try to estimate what the
future requirements for a product or SKU might be so that it is possible to
meet customer demand as closely as possible.

Forecasting, thus, helps in the inventory-holding decision process to find


answers to questions about what to stock, how much to stock and what
facilities are required.

There are several different approaches that can be used for forecasting:

● Judgemental methods – these are subjective assessments based on the


opinions of experts such as suppliers, purchasing, sales and marketing
personnel, and customers. Eg. Brainstorming, scenario planning and
Delphi studies.
● Experimental methods – these are used when there is no information
available on which a forecast can be based, for instance, new products.
Companies may set up a group of existing customers and determine
from them what the likely demand might be. Eg. Test marketing,
customer surveys and consumer panels.
● Causal methods – these are used where the demand for a product is
dependent on a number of other factors (promotions, price,
competitors’ plans, seasonality, weather, the state of the economy). The
main method used is regression analysis. Alternative approaches
include input–output models, simulation models and life cycle models.
● Projective methods or ‘time series’ models – these forecasting
techniques use historical demand data to identify any trends in demand
and project these into the future. They differ from the previous
approaches as no account of judgement or opinion is taken. They take
no direct account of future events that may affect the level of demand.
There are several different projective forecasting methods available,
and it is important to select the most appropriate alternative for
whatever demand is to be measured

Common Projective Forecasting Methods

Two of the most common methods of forecasting using time series are:
● moving average
● exponential smoothing.

Moving Average:
● The most simple is the moving average, which takes an average of
demand for a certain number of previous periods and uses this average
as the forecast of demand for the next period.

● The main problem with this method is that it ignores the age of the
observations as all are treated equally in the process of determining the
forecast demand level.

● It takes no account of changes over time, which is not really


representative of demand in most logistics systems – because things do
tend to change over time.

● This problem can be reduced by use of the weighted moving average,


which puts more emphasis on the use of the most recent data, so
reflecting actual demand more accurately.

Exponential Smoothing:
● A more sophisticated version of the weighted moving average is known
as exponential smoothing. This also gives recent weeks far more
weighting in the forecast, but each forecast is in fact a weighted
average of all prior 212 Procurement and Inventory Decisions
observations.

● The weighting process declines exponentially with the increasing age


of the observations, thus both emphasising the importance of the most
recent weeks but also taking account of the value of the earlier data.

● Forecasting methods such as exponential smoothing give a much faster


response to any change in demand trends than do methods such as the
moving average.

Diagram - Moving Average vs. Exponential Smoothing

Elements of a demand pattern:

● A trend line over several months or years. In the graph, the trend is
upward until the end of year 4, and then downward. It is this trend that
can be forecast by using methods such as exponential smoothing.
● A seasonal fluctuation. This is roughly the same – year in, year out. In
the graph, there is high demand in mid-year and low demand in the
early part of the year. A classic example for many products is the high
demand that occurs at Christmas.

● Random fluctuations that can occur at any time. These can be totally
random but they may be causal and thus identified by using some of the
qualitative methods described earlier. The classic example is – if it
rains, umbrella sales increase!

Diagram - Elements of a demand pattern


Advanced Projective Forecasting Methods

More advanced projective methods of forecasting include techniques such as:

● Double exponential smoothing (smooths random variation plus trend),


e.g. Holt-Winters method.
● Triple exponential smoothing (smooths random variation, trend and
seasonality), e.g. Winters method.
● Croston’s exponential smoothing (for intermittent demand – smooths
random variation and demand interval).
● Autoregressive integer moving average (fitted to time series data to
help in understanding the data and to predict future points in the series),
e.g. Box-Jenkins.
● Neural networks (artificial intelligence, where there is automated
learning from past experience).

Another important aspect of forecasting is measuring how accurate the


forecast is. In this regard, there are various measures.

For example, bias can be measured (eg the forecast always tends to be too
high or too low) and error can be measured.

Measures include:
● MAD: mean absolute deviation.
● MSE: mean square error.
● MPE: mean percentage error.
● MAPE: mean absolute percentage error
Approach to Demand Forecasting - Steps

It is sensible to adopt a very methodical approach to demand forecasting.

To achieve this, it is recommended that a number of key steps are used. These
can be summarised as follows:

1. Plan.
● Ensure from the outset that there is a clear plan for identifying and
using the most appropriate factors and methods of forecasting.
● Understand the key characteristics of the products in question and the
data that are available.
● Consider the different quantitative and qualitative methods that can be
used and select those that are relevant.
● If necessary and feasible, use a combination of different methods.
Identify ways of double-checking that the eventual results are
meaningful – it is unsafe merely to accept the results of a mechanical
analytical process.
● Forecasting at individual SKU level is a typical ‘bottom-up’ approach,
so check results with suitable ‘top-down’ information.

2. Check.
● Take care to review the base data for accuracy and anomalies.
● Poor data that is analysed will produce poor and worthless results.
● Where necessary, ‘clean’ the data and take out any abnormalities.

3. Categorise.
● A typical range of company products can display very different
characteristics.
● Thus, it is usually necessary to identify key differences at the outset and
group together products with similar characteristics.
● It is likely to be valid to use different forecasting methods for these
product groups.
● Use techniques such as Pareto analysis to help identify some of the
major differences: high versus low demand, high versus low value,
established products versus new products, etc.

4. Metrics.
● Use statistical techniques to aid the understanding of output and results
(standard deviation, mean absolute deviation, etc).
● There may be a number of relevant issues that can impact on the
interpretation of results: the size of the sample, the extent of the time
periods available.

5. Control.
● Any forecasting system that is adopted needs to be carefully controlled
and monitored because changes occur regularly: popular products go
out of fashion and technical products become obsolete.
● Control should be by exception, with tracking systems incorporated to
identify rogue products that do not fit the expected pattern of demand
and to highlight any other major discrepancies and changes.

You might also like