Exponential Smoothing (ES) Method
Posted: Sun Dec 22, 2024 8:53 am
Another very simple sales forecasting scheme is the exponential smoothing model, which is often used. In it, the estimated demand values depend on:
sales for the previous period;
demand forecast built for this period of time using some other method.
It is necessary to set the smoothing philippines whatsapp coefficient (α), taking into account these two factors. The larger α, the more the latest sales influence the forecast (from 0 to 10).
Forecast(t + 1) = (1 – α) x Forecast(t) + α x Sales(t).
The calculation is carried out with several values of α, after which the optimal one is selected. The method is considered to be working, but it should be remembered that the smoothing coefficient does not take into account trending or seasonal goods, etc. Based on this method, the Holt-Winters method was developed, which makes it possible to work with goods of different types.
How to achieve multiple growth in traffic and sales from your website?
Alexey Boyarkin
Dmitry Svistunov
Head of SEO and Development
Read more posts on my personal blog:
I have always been concerned about the issue of moving to a fundamentally new level. So that the indicators would grow not by 2 or 3 times, but by several orders of magnitude. From a thousand visits to ten thousand or from ten thousand to a hundred thousand, if we are talking about a website, for example.
And I know that such leaps are always the result of painstaking work in five areas:
Technical condition of the site.
SEO.
Collection of site semantics.
Creating useful content.
Working on conversion.
And at the same time, every manager needs an increase in sales and the number of applications from the site at the moment.
To get this growth, download our step-by-step template for increasing sales from the site:
Download template
Already downloaded
153116
Holt-Winters method
Since this formula is quite complex for a full analysis in an article, we will stop at looking at its logical chain.
Y^[t+h] = (a [t] + h x b [t]) x s [t - p + 1 + (h - 1) mod p],
Where:
h – for what period in the future we are counting;
Y^[t+h] – forecast for period number h;
p – seasonality period (for weekly – 7).
We have to build a forecast for the future period, which depends on many factors. What is inside this formula? Three main factors are distinguished - smoothing, trend and seasonality.
a[t] = [α * (Y[t] / s[t−p])] + (1 – α) * (a[t – 1] + b[t – 1]) – smoothing;
b[t] = β * (a[t] – a[t – 1]) + (1 – β) * b[t – 1] – trend;
s[t] = γ * (Y[t]/a[t]) + (1 – γ) * s[t – p] – seasonality.
Here α, β, γ are coefficients (from 0 to 1).
After calculating the seasonal and trend factors, finding exponential smoothing, and selecting the coefficients, a forecast value of demand for the future period appeared. This method is well suited for trending goods and seasonal products. Below is its implementation in an Excel table.
Holt-Winters method
The initial data contains monthly sales values. First, the seasonal factor is calculated using the average value method (by dividing actual sales by average sales for the period). As a result, we have seasonality coefficients. Then we calculated the trend coefficient and obtained the forecast value. Using this method, we take into account seasonality, trend, and exponential smoothing. The Holt-Winters method is suitable for those products that are characterized by stable sales and a pronounced trend.
sales for the previous period;
demand forecast built for this period of time using some other method.
It is necessary to set the smoothing philippines whatsapp coefficient (α), taking into account these two factors. The larger α, the more the latest sales influence the forecast (from 0 to 10).
Forecast(t + 1) = (1 – α) x Forecast(t) + α x Sales(t).
The calculation is carried out with several values of α, after which the optimal one is selected. The method is considered to be working, but it should be remembered that the smoothing coefficient does not take into account trending or seasonal goods, etc. Based on this method, the Holt-Winters method was developed, which makes it possible to work with goods of different types.
How to achieve multiple growth in traffic and sales from your website?
Alexey Boyarkin
Dmitry Svistunov
Head of SEO and Development
Read more posts on my personal blog:
I have always been concerned about the issue of moving to a fundamentally new level. So that the indicators would grow not by 2 or 3 times, but by several orders of magnitude. From a thousand visits to ten thousand or from ten thousand to a hundred thousand, if we are talking about a website, for example.
And I know that such leaps are always the result of painstaking work in five areas:
Technical condition of the site.
SEO.
Collection of site semantics.
Creating useful content.
Working on conversion.
And at the same time, every manager needs an increase in sales and the number of applications from the site at the moment.
To get this growth, download our step-by-step template for increasing sales from the site:
Download template
Already downloaded
153116
Holt-Winters method
Since this formula is quite complex for a full analysis in an article, we will stop at looking at its logical chain.
Y^[t+h] = (a [t] + h x b [t]) x s [t - p + 1 + (h - 1) mod p],
Where:
h – for what period in the future we are counting;
Y^[t+h] – forecast for period number h;
p – seasonality period (for weekly – 7).
We have to build a forecast for the future period, which depends on many factors. What is inside this formula? Three main factors are distinguished - smoothing, trend and seasonality.
a[t] = [α * (Y[t] / s[t−p])] + (1 – α) * (a[t – 1] + b[t – 1]) – smoothing;
b[t] = β * (a[t] – a[t – 1]) + (1 – β) * b[t – 1] – trend;
s[t] = γ * (Y[t]/a[t]) + (1 – γ) * s[t – p] – seasonality.
Here α, β, γ are coefficients (from 0 to 1).
After calculating the seasonal and trend factors, finding exponential smoothing, and selecting the coefficients, a forecast value of demand for the future period appeared. This method is well suited for trending goods and seasonal products. Below is its implementation in an Excel table.
Holt-Winters method
The initial data contains monthly sales values. First, the seasonal factor is calculated using the average value method (by dividing actual sales by average sales for the period). As a result, we have seasonality coefficients. Then we calculated the trend coefficient and obtained the forecast value. Using this method, we take into account seasonality, trend, and exponential smoothing. The Holt-Winters method is suitable for those products that are characterized by stable sales and a pronounced trend.