chapter 3 book: demand forecasting Flashcards
A demand forecast
the estimate of expected demand during a specified future period
Anticipated demand is derived from which two possible sources
actual customer orders
forecasts
which are the three types of uses for demand forecasts in operations?
(1) to help managers design the system
(2) to help them plan the medium-term use of the system
(3) to schedule the short- term use of the system
Two most important aspects of forecasts
(1) the expected level of demand
(2) the degree of accuracy that can be assigned to a forecast
collαborative planning, forecαsting, and replenishment (CPFR)
supply chain partners collaborating on the forecasting process
features common to all techniques and forecasts
- Forecasting techniques generally assume that the same underlying causal system that existed in the past will continue to exist in the future
- Forecasts are rarely perfect; actual results usually differ from predicted values
–> Allowances should be made for inaccuracies
- Forecasts for groups of items tend to be more accurate than forecasts for individual items, because forecasting errors among items in a group usually have a cancelling effect
- Forecast accuracy decreases the farther the forecasted time period is into the future
The forecasting horizon
the range of time periods we are forecasting for
that flexible business organization require which type of forecasting horizon?
require a shorter forecasting horizon
Elements of a Good Forecast
- The forecast should be timely
–> The forecasting horizon must cover the time necessary to implement possible changes so that its results can be used
- The forecast should be accurate， and the degree of accuracy of the forecast should be stated
- The forecasting method/software chosen should be reliable
- The forecast should be expressed in meaning of units
- The forecast should be in writing
- The forecasting technique should be simple to understand and use
- The forecast should be cost effective
Steps in the Forecasting Process
which are used on a continuing basis?
- Determine the purpose of the forecast
- Establish a forecasting horizon
- Gather and analyze relevant historical data
- Select a forecasting technique.
- Prepare the forecast
- Monitor the forecast
(only Steps 4 to 6 are used on a continuing basis)
the two general approaches to forecasting
judgmental
quantitative
judgmental approach to forecasting
consist mainly of subjective inputs, which may defy precise numerical description
rely on non quantitative analysis of historical data and/or analysis of subjective inputs obtained from various sources
ex: consumers (surveys), similar products (historical analogies), the sales staff, managers and executives, and panels of experts
quantitative approach to forecasting
involve either the use of a time series model to extend the historical pattern of data into the future
or
development of associative models that attempt to utilize causal variables to make a forecast
which forecasting approach permits the inclusion of soft information?
judgmental approach
long term forecasting uses which forecasting approach?
judgmental approach
Time series models
identify specific patterns in the data and project or extrapolate those patterns into the future
does not try to identify causes of the patterns
Associative models
use equations that consist of one or more explanatory variables that can be used to predict future demand for the variable of interest
Forecasting long-term demand typically involves what type of data?
annual data?
Forecasting medium-term demand typically involves what type of data?
monthly data
Forecasting short-term demand typically involves what type of data?
daily or weekly data
When introducing new products, redesigning existing products, and using sales promotions, which? forecasting is needed
judgmental forecasting
judgmental forecasts are based on what?
Executive Opinions
Sales Force Opinions
Consumer Surveys
Historical Analogies
Expert Opinions
Executive Opinions for forecasts
A small group of upper level managers may meet and collectively develop a forecast
often used as part of long-term strategic planing and new product development
advantage of executive opinions for forecasts
bringing together the considerable knowledge and talents of various managers
Sales Force Opinions for forecasts
direct contact with customers
often aware of any plans that the customers may be considering for the future, including the current level of customer inventory
Consumer Surveys for forecasts
enable corporations to sample consumer opinions
Historical Analogies for forecasts
Sometimes the demand for a similar product in the past, after some adjustment, can be used to forecast a new product’s demand
Expert Opinions for forecasts
we use the Delphi method
the Delphi method
circulating a series of questionnaires among experts
Responses are kept anonymous, which tends to encourage honest responses and reduces the risk that one person’s opinion will prevail
technological forecasting
an application of the Delphi method
assessing changes in technology and their impact on an organization
Often the goal is to predict when a certain event will occur
A time series
a time-ordered sequence of observations
taken at regular intervals over a period of time
(e.g., hourly, daily, weekly, monthly, quarterly, annually)
forecasting techniques based on time series data are made on which assumption?
on the assumption that future values of the series can be estimated from their own past values
the different time series patterns
Level (average)
Trend
Seasonality
Cycles
Irregular variations
Random variations
Level (average) pattern
a horizontal pattern of time series
Trend pattern
a persistent upward or downward movement in the data
Seasonality pattern
regular repeating wavelike variations generally related to factors such as the calendar, weather, or recurring events
cycles pattern
wavelike variations lasting more than one year
Irregular variations pattern
due to unusual one-time explainable circumstances not reflective of typical behaviour
Random variations pattern
residual variations that remain after all other behaviours have been accounted for (also called noise)
This randomness arises from the combined influence of many -perhaps a great many -relatively unimportant factors
cannot be reliably predicted
smoothed by time series techniques
the naïve method
can be used with:
a stable series (level or average with random variations)
seasonal variations
trend
the naïve forecast for stable series
the next forecast equals the previous period’s actual value
ex: if the demand for a product last week was 20 cases, the forecast for this week is 20 cases
the naïve forecast for data with trend
equal to the last value of the series plus or minus the difference between the last two values of the series
advantages of the naive method
virtually no cost
it is quick and easy to prepare
it is easily understandable
weakness of the naive method
the forecast just traces the actual data, with a lag of one period
it does not smooth the random variations out at all
Averaging Methods description
smooth variations in the data
smooth fluctuations in a time series because the individual highs and lows in the data offset each other when they are combined into an average
exhibits less variability than the original data
the three averaging techniques
- Moving average
- Weighted moving average
3 . Exponential smoothing
The moving average technique
averages a number of recent actual data values
uses the average as the forecast for the current period
updated as a new value becomes available.
The moving average technique formula
Ft = MAn = (EAi) / n
i = An index that corresponds to age of the period (i = t - 1: last period; i = t - 2: two periods back, . . )
basically just the average bro
n = Number of periods (data points) in the moving average
Ai = Actual value in period i
MAn = n period moving average
Ft =Forecast for this period (i.e., period t)
in the moving average technique, when is it more sensitive to new data changes?
the shorter the period (the fewer the data points)
in the moving average technique, when will the curve be smoother? what is the disadvantage of this?
averages based on more data points will be smoother
will be less responsive to “real” changes
The advantages of a moving average forecast
easy to calculate and easy to understand
A possible disadvantage of a moving average forecast
all values in the moving average forecast are weighted equally
weighted moving average technique
similar to a moving average
the difference is that it assigns larger weight to the most recent values in a time series in calculating a forecast
the weights sum to 1.00
the heaviest weights are assigned to the most recent values
involves trial and error
the advantage of a weighted moving average over a simple moving average
the weighted moving average is more reflective of the most recent observations
the Exponential Smoothing technique
a sophisticated weighted averaging method
a new forecast is based on the previous forecast plus a percentage of the difference between that forecast and the previous actual value
the Exponential Smoothing technique formula
Forecast = Previous forecast + α · (Previous actual - Previous forecast)
(Previous actual - Previous forecast) = forecast error
α: a proportion less than 1
the Exponential Smoothing technique, what is the importance of α
a smoothing constant
it determines the quickness of adjustment by forecast error
The closer its value is to zero, the slower the forecast will adjust by forecast error (i.e., the greater the smoothing)
the closer the value of αis to 1.0, the greater the responsiveness and
the less the smoothing
Selecting a smoothing constant in the Exponential Smoothing technique is a matter of what?
a matter of judgment or trial and error
adaptive (or variable response) exponential smoothing technique
Aversion of exponential smoothing where the smoothing constant is automatically modified in order to prevent large forecast errors from occurring
a feature that permits automatic modification of the smoothing constant so that the forecast errors do not become unacceptably large
one of the most widely used techniques in forecasting
Analysis of trend involves what?
developing an equation that will suitably describe the trend (assuming that trend is present in the data)
the linear trend equation
y = ax + b
a =
trend-adjusted exponential smoothing
Variation of exponential smoothing used when a time series exhibits trend
If a series exhibits trend, and exponential smoothing is used on it, the forecasts will all lag behind the trend: if the data are increasing, each forecast will be too low; if the data are decreasing, each forecast will be too high
The trend-adjusted forecast (TAF) for period t + 1 is composed of which two forecast?
TAFt+1 = St+ Tt
St = Smoothed series at the end of period t
St = TAFt + α · (At- TAFt)
At = actual value in period t
Tt = Smoothed trend at the end of period t
Tt =Tt-1 + ß(St - St-1 - Tt-1)
α and ß are smoothing constants
Seasonal variations
regularly repeating wavelike movements in series values that can be tied to recurring events, weather, or a calendar.
Most seasonal variations repeat annually, but can be shorter too
the two different models of seasonality
additive seasonality
multiplicative seasonality
additive seasonality model
expressed as a quantity
is added to or subtracted from the series average (or trend)
multiplicative seasonality model
seasonality is expressed as a proportion of the average (or trend) amount
it is then multiplied by the average (or trend) of the series
seasonal relatives
The seasonal proportions in the multiplicative model
why use seasonal relatives?
used first to deseasonalize the data
then later to incorporate seasonality in the forecast of deseasonalized data
deseasonalizing data
to remove the seasonal component from the data in order to get a clearer picture of the nonseasonal components
accomplished by dividing each data point by its seasonal relative
screw this one
vag
The complete steps of forecasting seasonal demand (time series decomposition)
- Compute the seasonal relatives.
- Deseasonalize the demand data.
- Fit a model to the deseasonalized demand data (e.g., moving average or trend).
- Forecast using this model (to obtain the deseasonalized forecasts) .
- Reseasonalize the deseasonalized forecasts.
how to compute seasonal relatives
We need to first compute the average (or trend) of all periods during the length of a repeating pattern
we use the centered moving average (CMA)
the centered moving average (CMA)
A moving average positioned at the centre of the data that were used to compute it
The implication is that the CMA is most representative of that point in the series
The number of periods needed in a centred moving average is equal to what?
the number of “seasons” involved
annual average method
A simpler method for finding seasonal relatives
involves finding the ratio of actual demand relative to average seasonal demand in that year and averaging the ratios across years
annual average method steps
- For each year, compute its total annual demand and average seasonal demand
- For each year, compute the ratio of the actual demand relative to average seasonal demand for each season
- For each season, average the ratios across years to get seasonal relatives
- Fit a linear trend to the total annual demand series
- Extend the model into the future to get next year’s total annual demand
- Divide the forecast of next year’s total annual demand by 4 to get the forecast of next year’s
average seasonal demand - For each season, multiply the forecast of next year’s average seasonal demand by the season’s seasonal relative
Cycles
wavelike movements, similar to seasonal variations, but of longer duration
two to six years between peaks
The most commonly used approach for cycles
Associative Models
Associative Models
Search for another variable that relates to, and leads, the variable of interest
rely on identification of related variables that can be used to predict values of the variable of interest
we want an equation that summarizes the effects of predictor variables on the variable of interest
the primary method of analysis for associative models
regression
regression
Technique for fitting a line to a set of points
predictor variables
Variables that can be used to predict values of the variable of interest
The simplest and most widely used form of regression
involves a linear relationship between two variables
linear regression
The objective of linear regression
to obtain an equation of a straight line that minimizes the sum of squared vertical deviations of data points (x, y) from the line
we use the least squares line
the least squares line
Minimizes the sum of the squared deviations around the line
y = a + bx
y=Predicted (dependent) variable
x =Predictor (independent) variable
b = Slope of the line
a = Value of y when x = 0 (i.e., the height of the line at the y intercept)
a = (Ey - bEx) / n or y - bx
b = (nExy) - (Ex)(Ey))
/
(n(Ex^2) - (Ex)^2)
n = Number of paired observations
correlation coefficient (r)
measures the strength of a relationship between two variables
can range between -1.00 to +1.00
+1.00 means both variables increase together 10%
-1.00 means that whenever a variable increases, the other will 100% decrease
correlation coefficient formula
r = (n · Exy - Ex · Ey)
/
((n · (Ex^2) - (Ex)^2)^(1/2) · (n · (Ey^2) - (Ey)^2)^(1/2))
the square of the correlation coefficient (r^2)
provides a measure of the proportion of variability in the values of y
the proportion is explained by the independent variable
can only range from 0 to 1.00
Models that involve more than one predictor require the use of which type of regression?
Multiple Regression
Forecast error
the difference between the value that actually occurs and the value that was forecasted for a given time period
Forecast error = Actual value- Forecast value
et = At - Ft
Positive errors result when?
when the forecast is too low relative to the actual value
Negative errors result when?
when the forecast is too high relative to the actual value
Accuracy of the forecasting process is measured using which three alternative forecast error summaries?
mean absolute deviation (MAD)
mean squared error (MSE)
mean absolute percent error (MAPE)
mean absolute deviation (MAD)
the average of absolute values of forecast error
(E|Actual - Forecast|) / n
mean squared error (MSE)
the average of squared forecast errors
(E(Actual - Forecast)^2) / n
mean absolute percent error (MAPE)
the average absolute percent forecast error
can be used to measure forecasting accuracy irrespective of the specific time series data used
(E [ |Actual - Forecast| / Actual] · 100) / n
why is it not meaningful to compare the MAD or MSE of two different time series variables?
Because MAD and MSE depend on the scale of data
for MAD, MSE, and MAPE, which accuracy percent is considered satisfactory?
above 70 %
bias
the sum of forecast errors
what does a persistently positive bias imply?
that forecasts frequently underestimate the actual values
what does a persistently negative bias imply?
implies that forecasts frequently overestimate the actual values
why do bias occur?
because either demand pattern has changed or the way forecasts are determined is flawed
Monitoring forecast errors is usually accomplished with what?
a control chart or a tracking signal
A control chart for forecast errors
a time series plot of forecast errors
the errors are compared to two predetermined values, or control limits
Forecast errors that fall outside either control limit signal that corrective action is needed
when using a control chart for forecast errors, when is corrective action necessary?
errors fall outside either control limit signal
Forecast eηors are randomly distributed around a mean of zero (i.e., there is no bias)
The distribution of forecast errors is normal
how do we choose the control limits for control charts?
usually chosen as a multiple of the standard deviation of forecast errors
The square root of MSE is used in practice as an estimate of the standard deviation of forecast errors
s = (MSE)^(1/2)
then we do 0 + or - 2s
Tracking Signal
An alternative measure to control the forecasting performance
sum of forecast errors divided by mean absolute
forecast error
TS = Ee / MAD
when using the tracking signal, when should the analyst study the model and make changes?
When TS values exceed 4 in absolute value
The two most important factors to consider when choosing a forecasting technique
cost and accuracy
the higher the accuracy, the higher or lower the cost?
the higher the cost
A reactive approach to forecast
views forecasts as probable descriptions of future demand
managers react to meet that demand,
a proactive approach to forecast
seeks to actively influence demand
requires either a causal model or a subjective assessment of the influence on demand
