1) CH3- Forecasting Flashcards
What is forecasting
process of predicting a future event
what is forecasting used for
production, inventory.personnel, facilities
WE USE THE PAST TO PREDICT THE FUTURE ALWAYS
WHY IS forecasting important
because businesses must make decisions, use resources, spend money before they know what the future holds
EX: call centre staffing, needs to forecast what? what is the pllaning horizon?
of calls per hour
1 day- 1 month
3 types of forecasting time horizions
1) short range : <3mos, up to 1 year
purchasing, job scheduling, workforce levels, job assignments, production levels
2) medium range: 3mos-3 years
sales and production planning, budget
3) long range: 3+ years
new product planning, facility location, R&D
Types of forecasts
economic
technological
demand
economic forecasts
address busienss cycle,
ifnlation rate, money cupply, housig starts
technologicla forecasts
predict rate of tech processes
impact development of new products
demand forecasts (focus)
predict sales of exisitng products and services
why does HR care about forecasts
hiring, training, lay offf workers
seven steps of forecasting
1) determine the use of the forecast
2) select the items to be forecasted
3) determine the time horizon of the forecast
4) select the forecasting models
5) gather data
6) make the forecast
7) validate and implement results
qualitiative methods of forecasting
subjective, used when no data available
new products and new tech! involves experience and intuition
quantiative methods of forecasting
used when situation is stable and historical data exists!
involves math techniques
5 examples of qualitiative methods of forecasting
1) jury of exec opinion (experts)
2) delphi and nominal group techniques (group consensus that arrives through an iterative process)
3) sales force ocmposite (Aggregatted from individual sales persons)
4) consumer market survey (Ask customers such as disney)
5) product life cycle curve (forecast life cycle stage A) INTRODUCTION B)GROWTH C) MATURITY D)DECLINE))
4 stages of product life cycle
introdcution
growth
maturity
decline
quantiative methods of forecasting example
1) time series forecasting (intrinsic) WE FOCUS ON THIS
2) casual methods (extrinsic)
time series forecasting (intrinsic)
relies on past data to predict the future
focus of this course
THIS IS WHAT WE USE
casual methods (Extrinsic)
regression, leading indicators, econometric models
includes use of external factors:
-forecasts for GDP based on housing starts
- exchange rates
- interest rates
what is selection criteria
important factors when determining which forecasting method to use
what are the selection criteria
why do we have these?
forecast horizon (daily-annual)
required accuracy (how much who shortage/excess cost???)
data availability (new porduct vs established one)
resources avaialble to make forecasts (people (knowledge), time, money)
b) because there are many techniques and we need to pick one
characterisitcs fo a good forecast
QUALITY AND COST- analysis
quality :
A) ACCURACY: Size (Spread) of forecast errors
B) BIAS: were predicitons consistently high or low? skew?
cost:
A) COST: cost of gaining additional data, more complicated models, and the value of an incrementally better forecast
B) CONSEQUENCES: what is the cost of not forecasting well? (Stocking too little or too much product/ under or over staffing)
5 characteristics of a good forecast
quality
cost
responsiveness
timeliness
simplicity
responsiveness and timeliness analysis
responsiveness: the forecast should reflect changes in market conditions quickly; outliers should be addressed
timelinesS: forecasts should be available for when they are needed
simplicty characteristic analysis
output should be easy to understad and interpet
what is time series forecasting
set of observations on a quantitatiev variable collected over time
ex: daily weekly monthly quarterly figures on sales/costs/profits
what does time series forecasting assue
factors influencing past and present will continue in the future
5 trends for analysing forecasting patterns graphically
1) no trend (demand is stable)
2) linear trend (postive or negative)
3) cyclical/seasonal (yearly)
4)trend and cyclical
5) little or no historical data
how do you forecast if no historical data (3 methods)
intutition/qualiitative methods
lok at similar products
do a market survey
statiionary time series forecasting
no discernible pattern in data
inspect the data plotted over time and try to find a pattern
forecasting using time series analysis- if data is stable/showing no patterns what can you use
smoothing methods
1) simple moving average
2) weighted moving average
3) exmplonential smoothing
forecastin gusin gtime serizes analysis- for data with patterns
trends (upward/downard)
cyclical seasonal
cyclaical/seasonal with trend (combo)
cyclical vs seasonal
they are similar (Seasonal is actually a cyclical pattern)
seasonal is related to season, cyclcilat related to something else
Naiive Approach
- is this a good forecasting technique
assume demand in next period is the same as demand in most recent period
if jan sales were 68 then feb is 68
NO!! NOT REALLY FORECASTING BRUH
why is naiive approach good? bad?
cost effective and efficienct
can be a good starting point
what is the “last period” tecnique?
naiive approach
forecasting notation, Ft At et
Ft= forecast of unkown value for some
period t in future (prediction)
At= actual observen known value for some period t in future
et= error between actual and forecasted value
et= At-Ft
3 forecasting smoothing methods
1) simple moving average (Average of several periods of data is the forecast [add up n prior period and divide by n for average])
2) weighted moving average (Allocate weights to prior periods)
3) exponential smoothing (choose alpha value for automativ weightings of a declining weighted moving average)
WEIGHTED MOVING AVG LOGIC
hwo to calcualte weighted moving average
Most recent data is most accurate (weighted more)
1) assign weights to the different months
2) multiply weight by data
3) add up all products
Simple exponential smoothing
basically you assign more weights to the recent data, less weight to the less recent data
our predicaiton for this month+alpha*error for this month
REARRANGE TO:
alphaActualT - (1-alpha)PredictedT
notes about exponential smoothing method
-you are not going to use any ‘n’ value
- to make prediction for march, you have to use the actual value from february and the prediction value from february as well
simple exponential smootghin formula
prediction for future month= alphaactual value + 1-alphapredicted value
all the values on the RHS come from the current month (come from march if you are predicting for april)
when do we use low values of alpha?
WHAT DOES IT MEAN IF ALPHA IS=0
when underlying average is fairly stablewh
alphaactual + (1-alpha)predicted
-> if alpha 0= we care about only the predicted, and ACTUAL demand is ignored
when do we use high alpha value
WHAT DEOS IT MEAN WHEN ALPHA=1
when underlying average is likely change
alphaactual + (1-alpha)predicted
-> if alpha 1= we care about only the last actual demand of the last month- > “LAST PERIOD TECHNIQUE”
how do you see if your forecasting method is good or not?
use the model on the past data to predict known data to see if it is useful ! how much error is there
what should a good forecast look like
scattered around zero!!! (This means the error is 0)
how do you select the model?
the one with least error
how to choose the forecasting method
use all of the models on past data
see if iti s accurate-> use them in the future
graph the error values-> close to zero then good!
Ex of annual, seasonal, monthly, weekly, daily product
christmas product, pumpking
plant ertilizer, posicles
banking, moving companies (end of month)
restaurants (depends on weekly promotions; wings on wednesday), bars, movies
traffice, fast food restaurants
Forecast formula
base*cyclical index
how do you forecast a cylical pattern? (3 steps)
1) use historical data, determine avg demand (set cyclical index=1.00)
CALCULATING THE BASE FOR THE FORECAST
2) use histroical data, caluclate cyclical endex for each period
3)Apply each periods cyclical index to forecast the demands
How do you forecast a cyclical trend
[Simple steps]
1) how much data points do you have (dont count average or any summative value) BE VERY CAREFUL!!!
2) calculate the base (Average for every period)
>average annual demand/12
=average monthly demand
3)calculate the cyclical index
(Average for period / Average of EVERY period)
> If below 1.0 it is below average
> if greater than 1 it is above average
4) Add all the indexes up and when you add them, you should get this to be equal to the number of periods!!!!! (THIS IS HOW U KNOW U DID IT WELL)
WHY IS forecasting so critical to a business
because businesses have toallocate resources before they know what demand is
how to calculate simple moviing average
1) Add up previous FORECASTED n period values
2) divide by n
what is a stationary time series
data that has no discernible pattern or trend
Exponential smoothing carries along (implicitly) all historic demand data but weights recent demand more heavily similar to an “exponential” WMA.
Exponential smoothing carries along (implicitly) all historic demand data but weights recent demand more heavily similar to an “exponential” WMA.
We call it simple eexponential smoothing because?
the alpha declines at an exponential rate
like a L shape
plotting error meaning
et=At-Ft
if it is close to zero then it is a good forecast
if positive= actual demand higher than predicted
if negative= actual demand less than presicted
if 0= no error
How do we measure how good a forecast is (3 ways)
1) Bias
2) MAD
3) Rise
Performance Measures: Bias
average error= sum of all errors/n
Performance Measures: MAD
mean absolute deviation=
sum of abs errors/n
Performance Measures: RMSE
RMSE= sqrt((sum of error values squared)/n)
another way to calculate RMSE?
Root of MSE
What does MAPE mean
mean absolute percent error (for comparng datasets)
how to compare and select forecasting models
1) identiy the forecasting methods and parameters
2) apply historical data and determine the errorcs (BIAS, MAD, RMSE)
3) select the best forecasting method (the one w lowest errors)
for BIAS MAD RMSE what are the values we want (no error?)
lowest values (2/3 is good hard to have lowest error in all 3)
what are non stationary models
models that have patterns
1) trend
2) seasonal
3) trends and seasonal combined
Forecasting w Patterns: Trend
upward or downard movement in a time series
use a base+trend model
Forecasting w Patterns: Seasonal
cyclical or repeating pattern in the data
use a “base* cyclical index “ tool
- avg demand=1
-above avg demand= >1
-below avg demand= <1
Forecasting w Patterns: Trens and seasonal
combine the “base +trend” and the “base*cyclical index”
what does trend mean?
how do you forecast trend?
it is persistent (positive or negative)
holts approach: Base+K(trend)
How to calculate cyclical pattern
**FIRST: COUNT HOW MANY PERIODS YOU HAVE!!
1) calc average demand for a year
avg demand monthly= [average annual demand/12]
(set this as your 1.00 since 1.00 is the avg in cyclical index)
2) use hisotircal data, calculate cyclical index for each period
cyclical index= [average of period/average demand monthly]
(if <1 lower than avg, if >1 higher than avg)
3) applly each periods cyclical index to future demand
-> find next years average monthly demand
-> use the previous years corresponding month to caclulate average monthly demand
(cyclical index*average monthly demand)
COMBO TIME! trend and cyclical index combo
Forecast= [Base + k(Trend)]*Cylical Index
BASE= avg quarterly attendance
Trend= given to you
Cyclicla INdex= calculated earlier
How to deal with outliers in data
keep if relevant to your period