1. Introduction:

“Forecasting future events is often like searching for a black cat in an unlit room, that may not even be there” Steven Davidson in The Crystal Ball

1.1 The requirement for Forecasting:

Increasingly competitive external pressures and complex environments have resulted in organisational survival becoming more dependant on the availability of accurate and timely information. Forecasting is a key technique used to reduce the uncertainty of the future and providing businesses with accurate information, regarding key economic and business variables that they require to make informed and reliable planning decisions.

1.2 Types of Forecasting Models:

In order to produce the most reliable forecast a suitable model needs to be chosen, with an understanding of the method’s applicability to the data and limitations within the operating environment. Alternative approaches to forecasting will initially be discussed and then a method for choosing an appropriate model for the specific data will be outlined.

Forecasting can be split between a qualitative (judgement) methods and quantitative (statistical) techniques. As stated by Makridakis, Wheelwright & Hyndman [1] both approaches are based on the same principle whereby existing patterns or relationships are identified and these are used as a foundation for prediction. The divergence lies within the documentation and processing of information prior to forecasting.

Qualitative procedures rely on subjective assessment and can centre on personal assessment whereas quantitative models base their predictions on objective analysis of data and making extrapolations from this.

Makredakis et al[2] provide a framework for matching the general characteristics of the forecasting situation with those of the various models. For our forecasting situation, constituting a medium-term time horizon, a single item, +/- 10% accuracy level and 20 years of historic data, a simple time series model seems appropriate (For further details of selection decision see section 2).

1.4 General Problems with Forecasting:

“Understanding the limitations of forecasting and setting realistic expectations … are central to making effective use of forecasts”[3]

As we have noted previously, forecasting is based on establishing patterns or relationships derived from historic data. Consequently, due to the dynamic nature of the economic and business environment, a fundamental disadvantage for modern forecasting is that these patterns and relationships are prone to dramatic change. Judgement of the magnitude and timing of these changes becomes a key factor in the accuracy of future prediction but is not within the scope of most forecasting models.

Academic debate has stemmed from Makridakis’s article “Forecasting: its role for planning and strategy” regarding the validity of long-term forecasting in today’s changing environment. It was argued by Makridakis that the historical data set chosen to extrapolate a forecast was a major determinate of its accuracy. It was therefore intimated that the key skill of forecasting was distinguishing long wave cycles (Kondradieff cycle) from long term trends and establishing an appropriate starting point for extrapolation. However, as highlighted by Grinyer[4] the importance of technological advancements and human’s ability to influence future events can not be ignored and could result in the long term equilibrium trend becoming less relevant as a basis for future forecasting. The consensus of this debate was that prediction of the general future direction is the best outcome that can be obtained from extrapolation of past figures in today’s environment. At least a general direction provides scenario planners with a starting point for their work.

2. Defining Time Series Models:

Time series methods aim to determine historical patterns and make the future predictions using a time-based extrapolation of the established patterns, assuming that these patterns recur over time. It is described as a black-box system which comprises purely of


A simple decomposition time series with linear trend line was chosen as our forecasting model. A moving average cyclical trend line was not used, even though it produced a closer match to historical data, because of the inherent limitations of being based solely on recent data points which was not as applicable to our data set.

The additive decomposition time series has three distinct components:

o Trend.
o Seasonal component.
o Random element.
D = T + S + R

The trend is the long term underlying movement in the variable being studied whereas the seasonality represents the variability that occurs in the series during a specified time period, usually one year. The random element is the difference between the actual data and that predicted