What is forecasting?

An attempt to determine in advance the most likely outcome of an uncertain variable

What is regular demand?

Items with a demand pattern that is nearly the same in the future or the demand entries must depend to some extent on the past values of a set of variables

What are the different categories of time horizons?

Long-term, medium-term & short-term

What is a long-term time horizon?

Forecasts span one to five years. Usually unreliable. Used for big decisions, such as whether to put an item on the market (Strategic level of decision making)

What is a medium-term horizon?

Forecasts span a few months to a year. Used for planning and budgeting (Tactical level of decision making)

What is a short-term horizon?

Forecasts span a few days to several weeks. Planning purchases, job scheduling etc. (Operational level of decision making)

Aggregate versus disaggregate forecasts

Aggregate forecasts are usually more accurate, as they tend to have a smaller standard deviation of error

What is the bullwhip effect?

Order variation is amplified as orders move farther from the end customer

Order variation is amplified as orders move farther from the end customer

bullwhip effect

What is it called when companies plan together?

Collaborative planning and forecasting

What are the steps in the forecasting process?

- Define objective
- Determine the time horizon of the forecast
- Select the forecasting method
- Data collection
- Perform the forecasting
- Control

What does Et mean?

Forecast error in period t

What does Ft mean?

Forecast in period t

What does Dt mean?

Demand in period t

What are the most popular error measures?

Mean squared error (MSE)

Mean absolute deviation (MAD)

Mean absolute percent error (MAPE)

What is MSE?

Mean squared error

Average of the squared differnces between the forecasted and observed values for periods 1, … , T

Punishes large mistakes more

What is MAD?

Mean absolute deviation

Average of the absoute valeus of the individual forecast errors for periods 1, … , T

Appropiate choice when cost of forecast error is proportional to size of error

What is MAPE?

Mean absolute percent error

Average of the absolute difference between forecasted and actual values, expressed as a percentage of the actual values.

Appropiate for more real results and when MAD or MSE are very big.

0-10% very good,

10-20% good,

20-30% moderate,

>30% poor

What are the qualitative forecasting methods?

Sales evaluation

Customer survey

Expert view

Delphi method

What are the quantitative forecastin methods?

Causal methods

-Linear-regression analysis

Time series analysis

What are causal methods?

Methods based on the hypothesis that future demand depend on the past or current values of some other variables.

What is a time series analysis?

method that presupposes that some features of the past demand time pattern will remain the same in the future.

Dependent variables v Independent variables

Dependent variables (y) depend on independent variables (x)

What is the least squares technique

A linear-regression analysis finds linear trens using LSE i.e. method that minimizes the sum of the squares of the vertical differences or deviations from the line to each of the actual observations.

Notes on least squares method

- Assumes a linear relationship
- Do not predict y far beyond the range
- Deviations should be normally distributed

What is a time series

Series of data indexed by time

Systematic component vs random compenent

Systematic component are the parts that we aim to understand. The random component refers to fluctuations in the data that cannot be explained by the underlying data patterns

What are the three basic demand patterns?

Trend (Tt), seasonal factor (St), residual variation (Rt)

When is Additive model / Multiplicative model most appropiate

Additive: if the magnitude of the seasonal fluctuations does not vary over time Dt = Tt + St+Rt

Multiplicative: if the amplitude of the variations in the seasonal pattern increases over time Dt=Tt*St*Rt

Linear regression for time series

Use periods (1, ... , T) and the normal linear regression stuff Used when Linear trend but no seasonality

Moving average

uses the average of the ‘r’ most recetn demand entries as the forecast for the first period ahead.

Forecasts after T stay the same

Used when demand has no observable trend or seasonality

Notes:

-Choice of r

-Cannot pick up trends (lag)

-Weight can be used to place more emphasis on more recent values

Simple exponential smoothing (Brown method)

Sophisticated weighted-moving average forecasting method with use of smoothing constant.

Forecasts after T stay the same

Used when no trend or seasonality

Notes:

-No forecast for first period ->Average Leave out for error calculation

-Alpha decides how important emphasis of past data is , can be optimized using Excel solver

-When optimized only relies on historical data

Trend-corrected exponential smoothing (Holt’s method)

Brown method extended with trend smoothing

Forecasts after T increase by last trend data

L1=D1 and T1=0

Initial values for L1 and T1 can be optimised

Used when linear trend but no seasonality

Trend- and seasonality- corrected exponential smoothing (Winters’ method)

Extension of Holt’s method including seasonality

Integer number of cycles

Split up between multiplicative variant and additive variant

Used when linear trend and seasonality

How do you normalize seasonal factor in Winter’s method?

E[S] = (S1 + .. + Sp)/p

Multiplicative:

E[S]=1

To normalize : Normalized St = St/E[S]

Additive :

E[S]=0

To normalize: Normalized St = St-E[S]

How to apply Winter’s method

- Compute T0, L0 then S1..Sp
- Compute Lt, Tt for every t e{1, .., p}
- Compute St, Lt, Tt for te{p+1, .. , T}
- Compute St for t e {T+1, .. T+p}
- Compute Ft for t e{1,..,T}
- Compute Ft for te{T+1,, … }