The various types of Time-series models are-
- Naive Method- It basically forecast value of the next period to
be same as the actual value of the present period. For example,
forecasted demand for Nth Period will be equal to actual demand at
(N-1)th period. It is usually used in small scale shops where error
in projection doesn't effect the profit much.
- Simple Average Method - In this, the forecast for the next
period is obtained by averaging all the previous historic values
i.e Demand for Nth period will be equal to average of 1 to (N-1)th
actual demands. This method is also used in small scale business
units where there is low fluctuation in demand and the error of
projection is not of big concern.
- Simple Moving Average - In this, the forecast for the next
period is equal to the average of a specified number of actual
values in the most recent observations. For example- For a 3 year
Moving average, the forecasted demand for Nth period will be the
average of last three actual demands i.e demand for (N-3), (N-2),
(N-1). This method is used in Mid segment businesses where there is
no sharp change in demand expected and a lower level forecast (low
customer service level) is acceptable.
- Weighted Moving Average - This is very similar to the Simple
Moving Average but each observation considered for the forecast is
assigned a different weight, with summation of all weights equal to
1. For Example- A 3 year moving average with weights (0.2,0.3,0.5)
with the most recent value getting highest observation is, given by
the sum of 0.2 time the (N-3)th actual value, 0.3 times the (N-2)th
actual value and 0.5 times the (N-1)th actual value. This method is
used in Mid segment business where there is no sharp change in
demand expected and a low forecast (low customer service level) is
acceptable.
- Exponential Smoothing is a technique which inculcates all the
previous actual and forecasted values when obtaining the forecast
for next period. In this technique, the weights of the the values
decrease from the most recent values to the older values. This
technique is obtained by using a very simple formula-> New
forecast = Last period’s forecast + Alpha*(Last period’s actual
demand – Last period’s forecast) where Alpha is the Smoothing
constant having value from 0 to 1. This technique is used in Large
business units where all historic data holds relevance and the
error needs to minimized to a great extent.
- Trend Projection - This method is a simplest form of linear
regression attempting to find a linear trend by using all the
actual values. Through this method, we are actually drawing a
straight line through all the historic data points in such a way
that the vertical deviation of the points from the trend line is
minimized. The equation of the trend line is usually -> Y = a +
bX, where Y is the forecasted value, X is the Observed value, b is
the statistically obtained slope value. This technique is usually
used where there is no sharp change in demand expected and there is
a observable & reliable trend in the data points.
- Seasonality Index Forecasting - In this method, we attempt to
sap out the periodic/seasonal behavior showed by the actual values
and use this characteristics to obtain the forecasted value. In
this technique, we usually use the seasonal index (which defines
the seasonality characteristics of the data) to de-seasonalize the
the data and then treat the then obtained data by using Trend
analysis. The forecast obtained from this Trend analysis is then
again multiplied with Seasonality Index to reintroduce the
seasonality characteristics and thus obtaining our actual
forecasts. This is usually used in Consumer Goods units where the
demand for a product exhibits a seasonal trend.
Associative Forecasting is a technique where we first determine
a causal relationship between the dependent variable and the
predictor and then utilize a regression method to obtain the
forecast for the next period. The regression techniques used in
defining this causal relationship can be both linear and non linear
regression model. For example, we are required to forecast sales in
a period, and we have data with respect to Price, Expenses, etc
with us. We will first determine the casual relationship between
Sales and all other independent variable and then identify the
strongest relationship out of them. Let us assume that the Sales
and the Expenses have a strong causal relationship, then we shall
utilize an appropriate regression model to obtain the sales
forecast for the next period using the data of the past
expenses.
This technique is a time consuming technique and we require to
have a great volume of data to train the regression model defining
the causal relationship. Thus, this technique is usually used in
the strategy formulation by big business units where a causal
relationship is commonly expected in business trends.
It is usually suggested that we use two or more type of
forecasting method to minimize the forecasting error. For example,
we can utilize causal, regression and seasonality methods to obtain
a more robust forecasting model and thus minimizing the forecasting
error to a greater extent.
Please Like & Provide your reviews in comment.
:-)