Forecasting Models: Trend, Holt’s Model and Seasonality

 Forecasting Models: Trend, Holt’s Model and Seasonality

Forecasting means estimating the future demand or sales of a product based on past data. In Supply Chain Management, forecasting is important because it helps organizations plan production, purchasing, inventory, warehousing, and distribution more effectively.

1. Constant Data

First, consider data where demand does not change significantly.

For example:

30, 31, 29, 30, 32, 29

Here, the demand stays around the same level. This is called constant data.

For constant data, we can use:

Simple Average

Weighted Average

Moving Average

Simple Exponential Smoothing

2. Data with Trend

Now consider:

26, 28, 29, 31, 32, 35

Here, demand is gradually increasing. This is called an increasing trend.

Using a simple average would not be very meaningful. The average of these values is approximately 30.16, while the latest demand is 35. Therefore, the next forecast is likely to be higher than 35.

So, we need a forecasting model that can capture the trend.

3. Linear Regression Model

One important method for forecasting trend-based data is Linear Regression.

The basic equation is:

Y = a + bt

Where:

Y = Forecasted demand

t = Time

a = Intercept

b = Slope or trend

Linear Regression finds a best-fit line that represents the overall pattern of the data.

For the data:

26, 28, 29, 31, 32, 35

the calculated values are:

a = 24.2667

b = 1.6857

Therefore:

Y = 24.2667 + 1.6857t

For the 7th period:

F₇ = 24.2667 + (1.6857 × 7)

F₇ ≈ 36.07

Therefore, the forecast for the 7th period is approximately 36 units.

4. Holt’s Model

One limitation of Linear Regression is that it gives equal importance to all the data points.

In forecasting, recent data may be more important than older data. Holt’s Model addresses this by using Exponential Smoothing.

The basic idea is:

Forecast = Level + Trend

or

Fₜ₊₁ = aₜ + bₜ

Where:

aₜ = Current level

bₜ = Current trend or slope

Holt’s Model gives more importance to recent observations. Therefore, it can respond better to recent changes in demand.

For the same example, Holt’s Model gives a 7th-period forecast of approximately:

36.32

The Linear Regression forecast was approximately 36.07.

The small difference occurs because Holt’s Model gives greater importance to recent data.

5. Seasonality

Some products have demand patterns that repeat during particular periods of the year. This is called seasonality.

Examples include:

Ice cream demand is usually higher in summer.

Air-conditioner demand is usually higher in summer.

Sweater demand is usually higher in winter.

Heater demand is usually higher in winter.

Therefore, demand may have both a trend and a seasonal pattern.

6. Seasonal Index

A Seasonal Index is used to measure the contribution of a particular season or quarter to the total demand.

For example, if annual demand is 157 and Q1 demand is 53:

Seasonal Index = 53 / 157

The same calculation can be performed for each quarter and for different years.

The seasonal indices from different years can then be averaged.

For example:

Q1 → 33.7%

Q2 → 14.0%

Q3 → 23.3%

Q4 → 29.3%

This tells us approximately how much of the annual demand occurs in each quarter.

7. Seasonal Forecasting

Suppose the annual demands of the previous three years are:

157, 173, 189

These values show an increasing trend. Using Linear Regression, the forecast for the fourth year is approximately:

205 units

Now, we can distribute this total demand across the four quarters using the seasonal indices.

Q1 = 205 × 0.337 = 69.08

Q2 = 205 × 0.140 = 28.70

Q3 = 205 × 0.233 = 47.77

Q4 = 205 × 0.293 = 60.06

Thus, we obtain the forecast for each quarter.

8. Winters’ Model

When demand has Level + Trend + Seasonality, a more advanced forecasting method is required.

This is where Winters’ Model is used.

Easy way to remember

Constant Data → Simple Exponential Smoothing

Trend Data → Linear Regression / Holt’s Model

Seasonal Data → Seasonal Index

Trend + Seasonality → Winters’ Model

The most important point in forecasting is to understand the behaviour of the data and select the appropriate forecasting model.

That is Exponential Smoothing in the simplest practical sense.

Support Our Work ☕

Use this link

Support me