Kaplan 3: Forecasting techniques & the Product Lifecycle Flashcards
Characteristic movements of time series
- Basic trend (long term)
General direction of the graph over a long interval of time after smoothing out short-term variations - Cyclical variations (not so long term but >1yr)
Trace cycle boom, decline, recession, recovery - Seasonal Variations (short term, not always seasons but patterns eg could be shifts)
Two models: additative & multiplicatitive - Irregular or Random variations (short term)
floods, strikes.
Unpredictable so can’t play a lge part in forcasting.
Can isolate by eliminating the other variations.
Important to remove any significant random vars from data before using them for forecasting (not in exam)
Time series: Calculation for seasonal variations
Time Series (Actual) = Trend + Seasonal Variation
TS = T+S
Time series: There are 3 ways of isolating the trend
- Drawing a scattergraph.
Quick & simple. Can extrapolate but not too far ahead. - Useing moving averages.
The effect of any seasonal variation can be eliminated to show the basic trend. (only works if used over the correct number of values … one full cycle) - Using linear regression.
Time Series: Isolate the trend using Moving averages remember..
Centring process:
If the number of values is even eg. 4
The average would be placed in the middle .. halfway between the 2nd & 3rd value rows.
In this case you have to ‘realign’ it by moving to a new column and calculating the average of each 2 values placing in the middle… now you have a figure opposite an actual figure and can work out the seasonal variation.
Times series: Disadvantages of moving averages
- Values at beginning & end of series are lost. Averages do not cover the complete period.
- The moving averages may generate cycles or other variations that were not present in the original data
- The averages are strongly affected by extreme values. (Sometimes a ‘weighted’ moving average is used , giving greater weight to the central items)
Time series: Isolate the trend using Linear regression.
what is the equation
(Regression analysis is a technique for estimating the line of best fit, given a series of data. It is essentially a statistical technique. Sometimes referred to as the ‘least squares method’)
y = a + bx
x is the independent variable
y is the dependent variable
a is the fixed element
b is the variable element
Approach to forecasting using Time series..
- Isolate/Extrapolate the trend to the appropriate future time.
- Adjust the extraplated trend value by the appropriate seasonal variation
Problems with forecasting using time series analysis
- Main problem is the Inherent weakness of extrapolation.
Past movement doesn’t guarantee future! - Seasonal adjustments used to find the forecast for the future are again based on historical figures which may already be out of date.
If there’s a large residual or random variation element this will make forecasts even less reliable
Population
all the items of information that the collector is interested in.
Eg if MA wanted to know the propotion of defective units produced by a machine in a day then the population would be all the units of product produced by the machine in a day
If info is required about a topic there are 2 main approaches..
Census
Examine every item in the population (rare in biz)
Sampling
Examine a sample of the population
(care must be taken when selecting a sample .. reliability of results depends on how unbiased the sample is)
Sampling Methods. 2 Main types each with 2 main sub-types.
- RANDOM.
- Best method for totally unbiased sample.
- Each item has equal chance of inclusion.
- Each item of pop must be known & assigned cons. no.
PURE random sampling (rare in practice)
Sample chosen using random tables or generator.
QUASI-RANDOM (Systematic; Stratified; Multi-stage)
Systematic - only first item random then every nth.
Stratified - if pop falls into distinct layers or groups.
Multi-stage - same but only random groups are incl.
- NON-RANDOM. :
- If random not cost-effective then non-random used.
- Not so accurate but info can still be useful.
QUOTA sampling
- Can be used when there are a number of diff groups in pop. eg men under 30, women over 30 etc.
- number of sample members from each group is determined then taken on non-random basis till quota is filled.
CLUSTER sampling
- Can be used where one or more areas of population are determined to be representative of pop as a whole therefore sample taken from that group alone.
- Eg buying habits of supermarket shoppers.. might just choose 3 different supermarkets in Birmingham as representative.
Main type Random
RANDOM > PURE Random
Every item in population must be known and have a consecutive number assigned.
Sample is chosen using RANDOM NUmbers taken from random number tables or a random number generator
Main type Random
RANDOM > QUASI-RANDOM SYSTEMIC sampling
RANDOM
Simpler method of random sampling than pure.
Again all items of pop must be known an assigned consecutive numbers.
FIRST item in sample is chosen using random number.
THEREAFTER every nth item.
(unlike pure where every selection is random)
Main type Random
Sub type 2.2. STRATIFIED sampling
RANDOM
Can be used if population falls into distinct layers or groups.
Sample is chosen from each group in proportion.
MULTI-STAGE sampling
RANDOM
Can be used if pop falls in fairly large groups or areas.
Initial stage select random groups or areas.
Next stage split into smaller groups from which again sample can be chosen randomly.
Can be done any number of times until final sample chosen.
