Marketers must regularly make decisions about future marketing activities, including alignment with overall organization strategy, marketing program investments and budgets, pricing, customer development, sales projections, and production forecasts. Historical sales performance helps reveal trends that, depending on anticipated business conditions, can directly impact the marketing plan.
Time series analysis is a useful method for using past quantitative data to predict future performance. Three popular methods are:
–naïve forecast;
–averaging forecasts; and
–exponential smoothing.
Naïve forecast
The naïve forecast assumes the next period’s demand will match the previous period. Forecasting requires that the variables are consistent in both the actual and forecast columns, as illustrated in Table 17.1 (i.e., use currency in both, or units).
Table 17.1: Naïve Forecast Chart
Period | Actual sales ($) | Forecast sales ($) |
January | 75 | |
February | 75 | 75 |
March | 90 | 75 |
April | 110 | 90 |
May | 120 | 110 |
June | 120 | 120 |
July | 150 | 120 |
August | 110 | 150 |
September | 100 | 110 |
Period | Actual sales ($) | Forecast sales ($) |
October | 90 | 100 |
November | 100 | 90 |
December | 130 | 100 |
Averaging forecasts
Averaging forecasts have several approaches–moving average and weighted moving average are two of the most common.
Moving average
Forecasters select a representative number of periods and calculate the average of those periods, with the result serving as the forecasted amount for the next period. As an example, consider a four month forecast period using the previous chart. In this case, the forecast is for total sales in the January–April timeframe divided by the number of periods (4), to arrive at May’s moving average. As illustrated in Table 17.2, May’s forecast sales are 88. The same process is repeated to determine June’s forecast sales (99), July’s (110), and so on.
Table 17.2: Moving Average Forecast
Period | Actual sales (thousands, $) | Forecast sales (thousands, $) |
January | 75 | |
February | 75 | |
March | 90 | |
April | 110 | |
May | 120 | 88 |
June | 120 | 99 |
July | 150 | 110 |
August | 110 | 125 |
September | 100 | 125 |
October | 90 | 120 |
November | 100 | 113 |
December | 130 | 103 |
The moving average forecast helps correct the simplistic assumptions of the naïve forecast since the previous period’s sales are unlikely to be perfectly repeatable in the next period. Moving average helps smooth over variations attributable to seasonal patterns. The moving average of sales performance based on the preceding months reduces the chance that any single month’s exceptional performance (good or bad) will unduly influence the next month’s forecast. However, recent sales data is usually considered more reliable than older data since it may be indicative of current market conditions. Moving average forecasts do not account for this since the impact of recent data is reduced due to the inclusion of older data in the average. The weighted moving average can help overcome this bias.
Weighted moving average
The weighted moving average (or “simple” weighted average) assigns weights to data in different periods with, generally speaking, more recent periods receiving a higher weighting because they are representative of current conditions and, therefore, seen as more influential. The sum total of all the weights equals 1, therefore, each weight is a fraction of 1. Continuing with the same example, assigning the lowest weight to the earliest month and the highest weight to the most recent as follows: .1, .2, .3, .4 (see Table 17.3):
May = January (75*.1) = February (75*.2) + March (90*.3) = April (110*.4) = 93.5
June = February (75*.1) + March (90*.2) + April (110*.3) + May (120*.4) = 106.5
July = March (90*.1) + April (110*.2) + May (120*.3) + June (120*.4) = 115
August = April (110*.1) + May (120*.2) + June (120*.3) + July (150*.4) = 131
September = May (120*.1) + June (120*.2) + July (150*.3) + August (110*.4) = 125
October = June (120*.1) + July (150*.2) + August (110*.3) + September (100*.4) = 115
November = July (150*.1) + August (110*.2) + September (110*.3) + October (90*.4) = 106
December = August (110*.1) + September (100*.2) + October (90*.3) + November (100*.4) = 98
Table 17.3: Weighted Moving Average Forecast
Exponential Smoothingii
Exponential smoothing is a more sophisticated approach to weighted moving average. It, too, is a popular forecasting technique used in computerized forecasting programs and wholesale and retail inventory ordering programs. Like the weighted moving average, exponential smoothing favors recent data over older data. A key difference, however, is the use of a “smoothing constant” called alpha, represented by ά. Alpha describes the level of smoothing deemed reasonable along with the speed of a company’s reaction to differences between forecasts versus actuals performance. As with weighted moving average, smoothing is a technique for reducing the impact of seasonality or more extreme variances from typical demand performance. It is always less than one and is based on the marketer’s experience and knowledge of what comprises a good response rate, combined with the characteristics of the product itself:
Where
Ft = new forecast
At = actual demand that occurred in the forecast period
Ft-1 = previous/most recent forecast
Forecasters begin the analysis with a previous period, building sequentially to arrive at the forecast for the period needed. Past data and/or the initial forecast from which to develop the analysis is required. Adapting the earlier table, we develop a forecast for April. To determine this, the forecasts for February and March must first be calculated. For February, we need to know Ft-1, the previous/most recent forecast (January, in this case). We assume it was 70 and that alpha is .6. The following result occurs for February (see Table 17.4):
An identical approach is used to determine the figures for March:
Table 17.4: Exponential Smoothing
Period | Actual sales (thousands, $) | Forecast sales (thousands, $) |
January | 75 | 70 |
February | 75 | 73 |
March | 90 | 74.20 |
April | 83.68 | |
May | ||
June | ||
July | ||
August | ||
September | ||
October | ||
November | ||
December |
Once the actual data for April is known, May can be forecasted, and the process continues as each month’s actual sales are included.
Marketers are accountable for developing forecasts to help determine market demand for their offerings. A time series forecast utilizes historical data and serves as a starting point for determining potential future performance, which would then affect marketing investment planning decisions. Time series forecasts help marketers observe and understand seasonal variation patterns in data as well as growth rate changes. However, marketers must be alert to the pros and cons of time series forecasts:
–They are never 100% reliable.
–Time series forecasts tend to be more accurate with shorter time frames (i.e. it is easier to predict tomorrow than it is next month, or next year).
–Time series analysis tends to assume that the future will be like the past.
–They tend to be more credible when based on longer data histories (i.e., using several months or years of data is better than several days).
–Newer data tends to be more reliable than older data and receives a higher weighting as a result.
When forecasting, marketers must determine if the sales trend is increasing, decreasing, or flat. Time series analysis can be helpful in answering basic trend questions as it may suggest emerging opportunities or, conversely, warning signs. But time series analysis is less useful for understanding and determining the causes that underlie trends. How do anomalous events such as external market disturbances (natural or man-made disasters), aggressive new marketing campaigns, or competitive behavior affect demand? What are the reasons for the seasonal variation? Time series analysis is a good first step toward developing a better forecast, but marketers must consider these other influences when developing their marketing plans.
The data for time series is found in historical business reports from finance, field sales, and production. As leaders in their respective organizations, marketers must turn on their proverbial antenna to detect weak signals in the market place as these, often more than historical patterns, will provide important guidance for factors that may impact their business planning assumptions.
Hossein Arsham, Time-Critical Decision Modeling and Analysis. Retrieved May 22, 2017 from http://home.ubalt.edu/ntsbarsh/stat-data/Forecast.htm#rgintroduction; Reference for Business, Forecasting. Retrieved May 22, 2017 from http://www.referenceforbusiness.com/management/Ex-Gov/Forecasting.html; Bob Namvar, “Economic Forecasting—How the Pros Predict the Future,” Graziadio Business Review 3, no. 1 (2000). Retrieved May 23, 2017 from http://gbr.pepperdine.edu/001/forecast.html; Qualitative Forecasting, Tutor2u. Retrieved May 22, 2017 from http://www.tutor2u.net/business/marketing/sales_forecasting.asp
iiG. Cachon and C. Terwiesch, Matching Supply with Demand: An Introduction to Operations Management, International Edition (New York: McGraw-Hill, 2006).