|Home | Training | Services | Testimonials | Contact | About Us|
|Demand Forecasting & Planning: The Trend Component|
The phrase Trend Forecasting means different things to different people. It may be used in a general way to refer to trends in fashion, lifestyle, colour or accessories. However, for those involved in forecasting future demand of the company's products for the purpose of balancing supply and demand over the medium-term, there is usually a need to forecast trends as an important component of the forecast. The same applies where the purpose of the forecast is to support financial projections including revenue and profitability. These considerations are important for all businesses and particularly so when tackling the forecasting task within a sales and operations planning process.
The most common approach to demand forecasting is to separate out the key forecasting components of trend, business cycle and seasonality from the sales history, project each of those components across the forecast horizon, then bring them all back together when calculating the final forecast. So forecasting trends is a part of the forecasting process, not its entirety. Separation of the various components was the basis of the traditional forecasting method called time series decomposition. Often the effect of business cycle, because it is so difficult to estimate, is combined with trend. Hence the phrase trend-cycle that appears in some text books.
If the company is heavily involved in promotions, this creates a further need to understand the effect of prior promotions and include future promotions as an additional component in the forecast. Most manufacturers of consumer packaged goods will understand this very well. Click here to read some thoughts on forecasting for FMCG products.
Trends due to known Causal Factors
Sometimes we may suspect or know that historical trends in sales of the company's products are cause and effect due to changes in factors such as population, relative pricing or economic indices. These factors driving the business are often referred to as causal factors and the mathematics to do with quantifying the effects of them is referred to as causal analysis or causal modelling. Such studies will sometimes be referred to as regression analysis because the mathematical technique that is used for the analysis is called regression. If the causal factor is to do with prices the phrases price elasticity or price sensitivity analysis may be used. If to do with weather variables, the phrase weather sensitivity may come into play.
If indeed it proves possible to carry out a satisfactory causal analysis this can be very helpful in putting together a process for demand forecasting. If the causal factor is to be used in future forecasting then a forecast for the causal factor has to be made as a starting point. If that is not possible it may be helpful simply to adjust the history to remove the effect of the causal factor in order to provide a level playing field for the creation of the forecast. This is sometimes done with weather effects, as a reliable weather forecast is not usually available more than a few days ahead.
Estimating trends when the cause is not known
If we can't identify sufficiently well the historical effects of specific causal factors it is still likely we will need to take trend into account as an important component of the future forecast. In a sense we are still taking a causal factor into account, but the causal factor is simply time period. We are now in the realms of what is often called Time Series Forecasting, where a time series is the set of sales data per time period.
Our forecasting training courses cover all the main approaches to demand forecasting and in some cases venture into stock policy and the setting of safety stocks.
One way of fitting a trend is the curve fitting approach, where a straight line or more sophisticated curved pattern is fitted to the sales history then projected into the future. For short-term forecasting a straight line is usually adequate, but for medium or long-term forecasting other curves may be more appropriate. The mathematical method that is used for this is usually the regression method referred to previously. The regression method is easiest with a straight line (linear) or with curves that can be reduced to a linear form with a simple mathematical transformation. Examples include the Linear, Polynomial, Exponential and Power Curve options that are included in Excel and some other software.
Forecasting with Seasonality
If product sales is subject to seasonal changes it is usually best to analyse the seasonal pattern and temporarily take this out of the history by way of a seasonal adjustment before tackling the matter of the trend projection. This applies to both causal analysis and time series approaches. Then take both seasonality and trend into account in calculating the final forecast. For further consideration of methods for seasonal analysis visit our page on forecasting with seasonality.
The process of seasonal decomposition is a good approach because it can be used with any forecasting method, however simple. Otherwise, as in many software products, one is drawn into very complicated methods such as Holt Winters that attempt to take trend and seasonality into account at the same time. Complicated forecasting techniques do not necessarily lead to the most accurate forecasts.
Moving Average and Exponential Smoothing
There are a number of other methods for trend forecasting and perhaps the most widely employed of these are moving averages and the family of methods called exponential smoothing. Moving average, as its name suggests, simply takes the average of a number of the most recent weeks or months of sales, then this estimate of the run-rate is used as a constant forecast going forward with no growth or decline i.e a flat or zero trend. Seasonality can be taken into account if it applies, as mentioned above
Exponential smoothing is similar to a moving average except that greater weight is given to the more recent time periods. In simple exponential smoothing the forward trend is flat, as in a moving average. Exponential smoothing forms a family of methods including some where trend forecasting is carried out. A fixed trend can be included, this having been estimated by a straight line fit over some or all of the data, or it can be based on a commercial judgement. Then there are more complicated methods that allow the rate of growth or decline to vary up or down during the historical simulation of forecast errors.
Forecasting in Excel
Forecasting with Excel has the advantage of giving complete flexibility in approach, but can be somewhat cumbersome in large or complex applications. Excel has a range of curve fitting options in its Fit Trendline tool and there is the Trend function for fitting a straight line and the Growth function for fitting an exponential curve. It provides a very basic causal analysis tool in the Data Analysis add-in. For some time there has been a Forecast function, also providing a straight line fit, but in Excel 2016 this has been extended. In Excel 2016 there is now Forecast.Linear and Forecast.ETS, the latter providing a rather complicated model that deals with both trend and seasonality. This always applies trend whether you want it or not, so this is not yet a complete basis for a forecasting system. However, it provides an additional tool that will be useful in some circumstances.
We use examples in Excel in our forecasting courses and there is the option of a course specifically on Excel forecasting.