Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 4

Forecast Models

When a series of consumption values is analyzed, certain patterns can usually be detected.
From these patterns it is then possible to differentiate between various forecast models:

 constant

 trend

 seasonal

 seasonal trend

A constant consumption flow applies if consumption values vary very little from a stable mean
value.
With a trend model, consumption values fall or rise constantly over a long period of time with
only occasional deviations.

If periodically recurring peak or low values which differ significantly from a stable mean value
are observed, it is a case of a seasonal consumption flow.

A seasonal trend consumption model is characterized by a continual increase or decrease of


the mean value.

If none of the above patterns can be detected in a series of past consumption values, then we
have an irregular consumption flow. In this case, it is better to go with moving average.

The models proposed for Popular are;

G – Constant model – when the consumption is pretty consistent

T – Trend Model – for linear, evenly rising or falling sales

S – Seasonal model – for periodically recurring sales patterns; note – the number of periods per
seasonal cycle must be maintained in the article master

X – Trend Seasonal Model – Combination of trend and seasonal model

 Formula for G;

Constant model

5 historical values (V1, V2, V3, V4 and V5)… V5 is the latest and V1 is the farthest

Forecast for periods F1, F2….

Smoothing parameter α = you can consider this in the calculation or ignore… the value can be
anything from 0.00 to 1.00

If the smoothing parameter is 0 then the formula is;

Forecast = Average (V1 + V2 + V3 + V4 + V5)

If the smoothing parameter is considered, then forecast value would be

Base value G2 = α * V2 + (1 – α) V1

G3 = α * V3 + (1 – α) G2

G4 = α * V4 + (1 – α) G3

G5 = α * V5 + (1 – α) G4
In this case the forecast value = F1 = F2 = Fn = G5

When we use smoothing parameters, it defines how strong the historical values would be
weighted, the further a consumption value in the past, the less it influences forecast value.

Hence if the smoothing parameter is small – consumption values far back in the past still
influence the forecast value and

If smoothing parameter is large – only consumption values in the immediate history affect the
forecast value

The generalized formula to calculate would be;

G (i) = α * V (i) + (1 – α) * G (i – 1)

Formula G is the base for the rest for the formulae.

 Formula for T

The difference between G and T forecast models is that in addition to the base value smoothing
parameter α, another smoothing factor β (trend value smoothing) is used.

Hence the formula would be

G (j) = α * V (j) + (1 – α) * G (j – 1)…. Base value G smoothing factor

T (j) = β * [G (j) – {G (j – 1)}] + (1 – β) * T (j – 1)

Hence the forecast value for period ‘k’ would be F (k) = G (n) + (k – n) * T (n)

Here J denoted a certain period with the application of smoothing parameter

 Formula for S
1. This is applicable to periodically occurring, non- calendar dependant time series
2. Number of periods per seasonal cycle should be known (this has to be part of the
article master); for example if there is a scenario that says ‘Back to School’
where the cycle would be one complete year, depending on the period indicator
maintained in the article master we need to define the number of periods in the
cycle (season). If the period indicator is Months then number of periods in the
season would be 12 and if the period indicator is Weeks, then the number of
periods in the cycle would be 52.

Formula – the difference between G and S forecast models is that in addition to the base value
smoothing parameter α, another smoothing parameter γ (smoothening the seasonal indices) is
used

Hence the formula would be;


G (j) = α * V (j) + (1 – α) * G (j – 1)…. Base value G smoothing factor

S (j) = γ * [V (j) / G (j)] + (1 – γ) * S (1 – p)

Here ‘p’ = periods per seasonal cycle

Hence the forecast value for period ‘k’ would be F (k) = S (k) * G (n) * S (k +/- x * p)

S = seasonal index, k = period, p = periods per seasonal cycle

 Formula for X

Here the system has to consider all the three smoothing factors α, β and γ

Hence the forecast value would be F (k) = [G (n) + (k – n) * T (n)] * S (k + / - x * p)

This is a mix of trend and season

You might also like