Business Forecasting Flashcards
Week 3 & 4
What is the interaction between companies/managers and uncertainty?
- All companies experience some levels of uncertainty
- Reduction in uncertainty helps make better decisions
- Managers wish to predict changes to allow for greater trend prediction so that adequate plans are in place
When is forecasting more accurate?
- Forecasts tend to be better in periods of economic stability, as there is more data to predict this period accurately
- Forecast horizons tend to be higher/more accurate when they are short-term, as little can occur in a smaller timeframe
What is the Hierarchy of Forecasts? Name the order of this
- Hierarchy of Forecasts shows the denominations of forecasts that can be predicted
- Move from national hierarchy to industry to individual firm to product
How best to select a forecasting technique?
- Cost associated with developing the forecasting model compared with its gains
- Complexity of the relationship that are forecasted
- Time period of forecast (LT/ST)
- Time between receiving information and using information
How can you assess the accuracy of a forecast (mathematically) ?
- Forecasting error = (Y - Ŷ)
- Root Mean Square Error (RMSE) = √[1/T Σ(Y - Ŷ)²
- Mean Absolute Deviation (MAD) = 1/T Σ |Y - Ŷ|
What are the two broad types of forecasting techniques?
- Qualitative (Looking for a direction of up/down)
- Quantitative (Uses accurate/exact estimates)
What are some advantages or disadvantages of qualitative forecasting?
- Flexible
- Shows early signals
- It can become complex to track all interactions
- Can be heavily biased
- There is a lack of test of accuracy
What are some examples of Expert Opinion forecasting?
- Personal Insight: An informed individual uses personal/company experience as a basis for expectations
- Panel Consensus: Based on several people’s opinions, the resulting forecast is an amalgamation of their opinions
What is the Delphi Method?
- Developed by RAND Corporation, where forecasts are derived from expert analytics
- Panel receive a series of questions individually and responses are analysed by an independent party
What are Surveys and Opinion Polls? What are some of the advantages and disadvantages?
- Surveys are interview/mailed questions that ask businesses/firms/Govt/individuals about plans
- Quick, cheap and give new ideas
- Can help appeal to consumer tastes
- Sample bias/biased questions
- Could lie
What is a Market experiment? What can be a disadvantage of this?
- Select a test market to design a strategy
- Experiment is designed to test reactions, prices and products
- Risk of losing consumers- so can’t control all factors
What are Barometric Indicators? Give an example of how these can help this
- They are designed to alert businesses about general economic conditions
- Lagging and Leading indicators are slightly out of sync- so leading indicators can be useful for forecasting (but these are still out of sync)
- Confidence board have a list of indicators
- Average working week can show manufacturing output
How can you best choose a Leading Indicator [qualities]?
- Must be accurate
- Must provide adequate lead time
- The lead time should be consistent
- There should be logic in why it predicts the outcome
- The cost/time necessary for data collection should be minimised
How can these indicators be interpreted?
- Composite Indicies [weightedness]
- Diffusion Indicies [proportionality]
What is the difference between Composite and Diffusion Indices?
- Composite Indices: Weighted Average of individual indicators and interpret them in terms of % change
- Diffusion Indices: Measure the individual time series that increases one month to the next
What are some disadvantages of using Barometric Indicators?
- Non-perfect prediction, variability in Lead times
- High variability of indicators month-to-month
- Doesn’t imply the magnitude of changes
- Inconsistencies in component indicators
What are some advantages or disadvantages of quantitative forecasting?
- This can illustrate the organised or behavioural relationship
- Easy to test accuracy (reliability) of forecast
- Relies on past data -> cannot provide us real time
- Data mining of some information
What are some quantitative forecasting techniques?
- Deterministic trend analysis: Time Series Data are a sequence of the values of an economic variable at different points (assuming that future patterns in a time series might project past patterns)
- Smoothing Techniques: Use average of the past to predict the future
- Econometric Modelling: Using Time series/Cross-sectional/ both
- Input-Output Analysis: Interrelationship between different sectors/industry
What are the components of a time series?
- Secular trends: Long-run trends in economic data
- Cyclical components: Business Cycles -> major expansions
- Seasonal effects: Seasonal variation -> fairly constant
- Random fluctuations: Unpredictable random factors
How do you model a time series?
- Using past data to project future
- Yt is the actual variable, Ŷt is the forecast
- Ŷt = f(Yt-1, Yt-2, Yt-3…)
What are the criteria for prediction accuracy [mathematics] ?
- MSE = 1/T Σ(Y - Ŷ)² ; RMSE =√MSE
- The lower MSE/MAD/RMSE
What are the elementary time-series models?
- (1) Ŷt+1 = Yt
- Simplest method, works where there is no trend
- Need to collect the data quickly to make good forecast
- (2) Ŷt+1 = Yt + (Yt - Yt-1)
- Additional adjustment to include changes
- Model predicts better if there are +ve/-ve trends
What are Secular trends? What are the different kinds of trajectory?
- By plotting data, we will see a trajectory from the line of best fits
- If Linear, the equation is Y = α + βt
- If constant growth, the equation is Y = Y0 (1+g)ᵗ
- If declined growth, Y = e ^ (β1 - β2/t)
- Parameters β, β1, β2, g and α can be found through OLS
How can you adjust for seasonality (name the two methods)?
- Ratio-to-trend method
- Season dummy variable method
Detail the plan of ratio:trend method
- Find the trend forecast [linear trend => Ŷ = α + βt ]
- Divide the actual values by trend forecast values
- Using these ratios, compute an average ratio per month/quarter
- Multiply trend forecast values each month by seasonality to give a seasonally adjusted forecasting
Detail the Seasonal Dummy method
- Incorporate seasonality into linear trends
Ŷ = β0 + β1t + β2D1t + β3D2t + β4D3t … where D1t = 1 for Q1, 0 if not etc. - Estimate parameters via OLS/past data
- For N periods, you need N-1 dummies to avoid the ‘dummy trap’
What are smoothing techniques? Name the two methods
- Attempt to average out random variables
- Works best when a data series changes slowly
- Can use moving average or first-order exponential smoothing
What is the formula for Moving Average smoothing technique?
- Ŷt+1 = (ΣYt) / n
- To choose the correct N, RMSE should be minimised
- Each past observation has the same weight of 1/N
What is the formula for First-Order exponential smoothing technique? What is the fundamental difference between this and the moving average method?
- FOES puts a weight on observations- giving greater weight to more recent observations
- Ŷt+1 = Ŷt + w (Yt - Ŷt)
- OR Ŷt = wYt-1 + (1-w) Ŷt-1
- Combining the two equations, we get:
Ŷt+1 = wYt + w(1-w)Yt-1 + w(1-w)² - Choose a w that minimises RMSE
What is the process of constructing an econometric model?
- Specify the Model
- Identify Variables
- Collect data
- Estimate the parameters of the model (α, β …)
- Develop a forecast based on parameters
- NOTE : ε~N (0, σ²)
What are some issues with econometric models?
- Simultaneity/identification problem- some independent variables can be endogenous
- Multicollinearity- independent variables may be highly related
- Autocorrelation- Error terms may have a pattern
- Heteroscedasticity- Error terms may have a non-constant σ²
