Business Forecasting Topic 1 Flashcards
why forecast?
- uncertainty in future
- mitigate impact of uncertainty
poor forecasts lead to
3 consequences
- too much stock
- unnecessarily poor service
- over or under
Finding a forecast
- underlying pattern in past data
- not explicit
components of observations
- systematic element (signal)
- random element (noise)
Actual value formulae
signal + noise
noise in forecasts
- irregular, random events
- filter out noise
- isolate systematic pattern (forecast future development)
Quantitative forecasting techniques
- assume of continuity (past pattern in signal = continue to future)
- time series extrapolation
- explanatory
assumption of continuity
use quantitative methods when….
- sufficient past data
- info quantified
- valid assumption of continuity
time series forecasts
univariate method
- analyse and extrapolate past pattern
- no past data (explanatory variables) - expensive
- no idea what influences variable or explaining behaviour
- no expertise to have elaborate model, not justified by extra accuracy “might” yield
- cheap and simple
explanatory forecasts
linking variable with other variables (independent & dependent variables)
- factor explains past through relationships with other variable
judgmental forecasts
- subjective using expertise
1- little/no relevant past data
2- forecaster knowledgable about unique event
- most widely used
five ways of combining forecasts from different methods
- statistical and judgment
1. apply judgment adj to stats forecasts
2. simple av forecast from diff methods
3. weighted av of forcast from diff methods
4. bayesian - judgement estimate incorporated in stats
5. rule based - weighted av from methods used weights based on particular conditions at time of forecast
Basic Steps in the forecasting task
- Define the problem
- Gather information
- Exploratory data analysis
- Choose method
- Evaluate chosen forecasting methods
Explanatory data analysis (preliminary)
graph variable
- consistency of patterns
- trend and seasonality
- economic cycles/outliers are present
Overfitting
model fits past data well doesn’t guarantee accurate forecasts
- method is seeing systematic patterns in the noise