2. Introduction
We make forecast everyday in our daily life
E.g. i. When should I leave to catch the flight
ii. How much food and drink will be required
for the party
To make this 2 fundamental things are considered
a. Current factors or conditions
b. Past experience in similar situations
3. Features
1. Underlying assumption that the same causal system that existed in past will
continue to exist in future.
2. Forecasting is never perfect , you never know what factors will impinge more.
3. Forecasting for a group of items tends to more accurate
4. As the time-horizon increases the forecasting
accuracy tends to decrease.
4. Steps in Forecasting
1. Determine the purpose of forecast : How will it be used, the amount of detail
required.
2. Establish a time horizon : One must keep in mind the time interval, keeping in
mind the accuracy.
3. Select a technique
4. Obtain,clean and analyze appropriate data: Clean the data of
outliers, any repetitive data.
5. Make the forecast
6. Monitor the forecast: To judge weather the forecast model is
6. Based on Time Series Data
Future values are estimated based on the past data and no attempt is made to
identify variables which influence the series
Following type of pattern may be observed
Trends : Long-term upward or downward movement
Seasonal : Weekly footfall in cinema hall, “Seasonal” buying of certain goods
Cycles : Wavelike variation lasting for more than an year
Variation around an average :
7. Methods
Naive Method : A forecast for any period that equals the previous period’s
actual value.
Moving Average Method : Technique that averages a number of recent actual
values, updated as new values become available
F1 = (n1 + n2 + n3 + ...) / n
Responsiveness, a moving average with less data points is more responsive
8. Weighted Moving average: More recent values in a series are given more
importance
Ft+1 = wt1(Dt) + wt2(Dt-1) + wt3(Dt-2)
Advantage over simple moving is forecast is reflective of more recent occurrences
Exponential Smoothing : A weighted average method based on previous
forecast plus a percentage of forecast error.
9. Next Forecast = Previous Forecast + k(Actual-Previous Forecast)