Forecasting Flashcards
forecasting
- most important business function, critical to meeting demand
- all other business decisions are based on the forecasts of the future
- poor forecasting leads to incorrect business decisions and leaves the company unprepared to meet future demand
- risk of over forecasting -> having too much of an expensive product that will quickly become obsolete
common features of forecasting models
- forecasts are rarely perfect -> impossible to make perfect predictions because of uncertainty. Goal is to minimize the average error over time
- forecasts work better for group items, rather than individual items because the individual high and low values cancel each other out
- forecasts are more accurate for shorter time horizons than longer time horizons, because the degree of uncertainty increase as time horizons increase
steps in forecasting
- identify what to forecast
- evaluate and analyze data
- select and test forecasting model
- generate the forecast
- monitor forecast accuracy
types of forecasting models
- Qualitative method -> judgmental method
- forecasts made subjectively by forecaster
- educated guesses by experts based on intuition, knowledge experience
- often biased by motivation, mood or conviction o forecaster
Pro: can incorporate last-minute information - Quantitative method -> based on mathematical modeling
- objective (does not suffer from biases) and consistent
Cons: require quantifiable data (not always available) and selection of appropriate model
effect of inaccurate forecasts ?
one factor that significantly impacts sales?
- billions in missed sales or excess inventory
- the weather -> new companies help companies use weather data to predict consumer behavior and manage weather risk. helps company move inventory to right places where customers will need them
common qualitative methods
- executive opinions -> managers gather to come up with forecasts. Good for strategic or new product forecasting. One person’s opinion might dominate the rest
- market research -> uses surveys and questionnaires to identify customer preferences and identify new product ideas. Good determinant of customer preference. Hard to develop a good questionnaire
- delphi method -> seeks to develop consensus among a group of experts. Excellent for forecasting long-term product demand, technological changes, or scientific advances. No one person can dominate. But time-consuming to develop
common quantitative methods
- time series models -> assume all information needed to forecast is in the time series of data (series of observation taken at regualr intervals over specified period of time). Believes that we can generate a forecast based on patterns in the data -> trends, seasonality and cycle -> easier than causal, just as accurate and is simplle to use, and generates forecasts quicker
EX: naive method, simple moving average, exponential smoothing - causal models (associative models) -> assume that forecast is related to other variables in the environments -> forecaster figure out how these variables are related in mathematical terms
complex, takes longer to forecast because it requires model building
4 basic patterns of time series models
- level/ horizontal pattern -> data fluctuates around a constant mean. Simplest and easiest to predict. Usually,
products in their mature stage that sales do not increase or decrease over time - trend -> data exhibits a nonlinear increasing or decreasing pattern over time
- seasonality -> pattern that regularly repeats itself and is of constant length
- cycles -> patterns created by economic fluctuations (recessions, inlations, life cycle of product). different from seasonal pattern because it varies in length and magnitude. Most difficult to forecast
forecasting level pattern
Naive method -> simplest forecasting model
- Assumes next period’s forecast is equal to the current period’s actual
Simple mean or average -> forecast made by taking an average of all data
- only good for level data pattern
Simple moving average -> forecast made by taking an average of actual data of the n most recent periods
- as new data becomes available, old gets dropped
- only good for level data pattern
forecasts with simple moving average
if alpha is large, forecasts are very responsive
if alpha is small, forecasts remain stable
- the smaller the number of observations in the moving average, the more RESPONSIVE the forecast is to changes in demand
BUT it is more subject to the random changes in the data (if data contains a lot of randomness, high responsiveness leads to greater errors) - the larger the number of observations in the moving average, the LESS responsive the forecast is to changes in demand, but also to the randomness. These forecasts are more stable.
selection is based on the characteristics of the data
exponential smoothing model
forecasting model that uses a sophisticated weighted average procedure you need: current period forecast current period actual value smoothing coefficient (alpha)
selecting alpha: you can choose to place more weight on either the current periods actual or the forecast
low alphas generate stable forecasts because its doesn’t place that much weight on the current periods actual demand
high alphas place alot of weight on the current periods actual demand and can be influenced by random variations
forecasting trend (2 ways)
trend-adjusted exponential smoothing
linear trend
trend-adjusted exponential smoothing
exponential smoothing model used for data that exhibit a trend
- uses 3 equations:
1 (smooths out level of the series) -> based on actual data and the forecast of the current period, without considering that the trend continues in the next period
2 (smooths out trend) -> based on the trend modeled in the current period and the expected level difference between the next and current period
3 (generate forecast by adding up the findings of the first 2 equations)
—> 2 smoothing coefficients alpha (0-1) and beta (0.1-0.2)
forecasting seasonality
- any regularly repeating pattern is a seasonal pattern (sales of turkey before thanksgiving)
- seasonal index is the percentage by which the value for each season is above or below the mean
Ex: enrollment for fall semester is 1.30 if the mean -> fall enrollment is 30% above the average
forecasting procedure:
1. calculate average demand per season (Total annual demand/ number of seasons)
- Determine seasonal index for every season of every available year (Actual demand for each season/ average demand per season)
- determine the average seasonal index for every season (add up seasonal index values for that season and divide by number of years)
- use any forecasting method to compute deseasonalized demand for next year
- multiply the forecasts for every season by the average seasonal index
linear regression
In linear regression the variable being forecast (dependent variable) is related to some other variable (independent variable=) in a straight line way
dependent variable Y = forecast
Y= a + bX
linear regression selects parameters a and b, which define a straight line that minimizes the sum of the squared errors, or deviations from the line (leas - squares straightt line)
