Basic Sales Forecasting Lingo
You need to get a handle on the specialized terminology used in sales forecasting for a couple very practical reasons. One is that you may be asked to explain your forecasts to your boss or in a meeting of, for example, sales managers.
Another good reason is that Excel uses many of these terms, as do other programs, and figuring out what’s going on is a lot easier if you know what the terms mean.
Autoregressive integrated moving averages (ARIMA)
If you’re going to do forecasting, some smart aleck will eventually ask you if you used autoregressive integrated moving averages (ARIMA), and you should know how to reply. ARIMA is in part a forecasting method, and also a way of evaluating your baseline so that you can get quantitative evidence that supports using a regression approach, a moving-average approach, or a combination of both. Unless you really take to this forecasting stuff, you’ll usually do just fine without it, even though it’s an excellent, if complex, diagnostic tool.
By the way, your answer to the smart aleck should be, “No. I’ve been working with this baseline for so long now that I know I get my best results with exponential smoothing. Which, as you know, is one of the forms that ARIMA can take.”
A baseline is a sequence of data arranged in chronological order.Some examples of baselines include total monthly revenues from January 2010 through December 2015, number of units sold weekly from January 1, 2015, through December 31, 2016, and total quarterly revenues from Q1 2007 through Q4 2016. Data arranged like this is sometimes called a time series.
A correlation coefficient expresses how strongly two variables are related. Its possible values range from –1.0 to +1.0, but in practice you never find correlations so extreme. The closer a correlation coefficient is to +/–1.0, the stronger the relationship between the two variables. A correlation of 0.0 means no relationship. So, you might find a correlation of +0.7 (fairly strong) between the number of sales reps you have and the total revenue they bring in: The greater the number of reps, the more that gets sold. And you might find a correlation of –0.1 (quite weak) between how much a rep sells and his telephone number.
A special type of correlation is the autocorrelation, which calculates the strength of the relationship between one observation in a baseline and an earlier observation (often, but not always, the relationship between two consecutive observations). The autocorrelation tells you the strength of the relationship between what came before and what came after. This in turn helps you decide what kind of forecasting technique to use. Here’s an example of how to calculate an autocorrelation that might make the concept a little clearer:
This Excel formula uses the CORREL function to show how strong (or how weak) a relationship there is between whatever values are in A2:A50 and those in A1:A49. The most useful autocorrelations involve baselines that are sorted in chronological order. (This sort of autocorrelation is not quite the same as the autocorrelations calculated in ARIMA models.)
A cycle is similar to a seasonal pattern, but you don’t consider it in the same way as you do seasonality. The upswing might span several years, and the downswing might do the same. Furthermore, one full cycle might take four years to complete, and the next one just two years. A good example is the business cycle: Recessions chase booms, and you never know just how long each is going to last. In contrast, yearly seasons have the same length, or nearly so.
The damping factor is a fraction between 0.0 and 1.0 that you use in exponential smoothing to determine how much of the error in the prior forecast will be used in calculating the next forecast.
Actually, the use of the term damping factor is a little unusual. Most texts on exponential smoothing refer to the smoothing constant. The damping factor is 1.0 minus the smoothing constant. It really doesn’t matter which term you use; you merely adjust the formula accordingly.
Stupid term, even if technically accurate. Using exponential smoothing, you compare your prior forecast to the prior actual (in this context, an actual is the sales result that Accounting tells you — after the fact — that you generated). Then you use the error — that is, the difference between the prior forecast and the prior actual — to adjust the next forecast and, you hope, make it more accurate than if you hadn’t taken the prior error into account.
The forecast period is the length of time that’s represented by each observation in your baseline. The term is used because your forecast usually represents the same length of time as each baseline observation. If your baseline consists of monthly sales revenues, your forecast is usually for the upcoming month. If the baseline consists of quarterly sales, your forecast is usually for the next quarter. Using the regression approach, you can make forecasts farther into the future than just one forecast period, but the farther your forecast gets from the most recent actual observation, the thinner the ice.
You’ve probably run into the concept of moving averages somewhere along the line. The idea is that averaging causes noise in the baseline to cancel out, leaving you with a better idea of the signal (what’s really going on over time, unsullied by the inevitable random errors). It’s an average because it’s the average of some number of consecutive observations, such as the average of the sales in January, February, and March. It’s moving because the time periods that are averaged move forward in time — so, the first moving average could include January, February, and March; the second moving average could include February, March, and April; and so on.
There’s no requirement that each moving average include three values — it could be two, or four, or five, or conceivably even more.
You generally find this term in use when you’re forecasting with regression. The predictor variable is the variable you use to estimate a future value of the variable you want to forecast. For example, you may find a dependable relationship between unit sales price and sales volume. If you know how much your company intends to charge per unit during the next quarter, you can use that relationship to forecast the sales volume for next quarter. In this example, unit sales price is the predictor variable.
If you use the regression approach to sales forecasting, it’s because you’ve found a dependable relationship between sales revenues and one or more predictor variables. You use that relationship, plus your knowledge of future values of the predictor variables, to create your forecast.
How would you know those future values of the predictor variables? If you’re going to use unit price as a predictor, one good way is to find out from Product Management how much it intends to charge per unit during each of the next, say, four quarters. Another way involves dates: It’s entirely possible, and even common, to use dates (such as months within years) as a predictor variable.
During the span of a year, your baseline might rise and fall on a seasonal basis. Perhaps you sell a product whose sales rise during warm weather and fall during cold. If you can see roughly the same pattern occur within each year over a several-year period, you know you’re looking at seasonality. You can take advantage of that knowledge to improve your forecasts. It’s useful to distinguish seasons from cycles. You never know how long a given cycle will last. But each of four seasons in a year is three months long.
A trend is the tendency of the level of a baseline to rise or fall over time. A rising revenue trend is, of course, good news for sales reps and sales management, to say nothing of the rest of the company. A falling baseline of sales, although seldom good news, can inform Marketing and Product Management that they need to make and act on some decisions, perhaps painful ones. Regardless of the direction of the trend, the fact that a trend exists can cause problems for your forecasts in some contexts — but there are ways of dealing with those problems.