How can you describe Demand Forecasting?
Refers to the process of predicting the future demand for the company products by using all the available information. It is the anticipation of the demand of the product in the future.
Also it is the required quantity (demand) and can be different sales.
Which are the main methods of demand forecasting? mention some examples of each method.
- Time Series Approach [ARIMA, Moving average, Exponential smoothing, Decomposition]
- Relational/Causal Approach [Regression, Econometrics]
- Qualitative Approach [Sales force, Expert panels, Future scenarios/analogies, Marketing]
Mention Issues of Demand forecasting
- Product Level
- Planning horizon
- geographical aggregation
- influencing variables
- required accuracy
What is Time Series of the Demand?
Note: Different from the “Times Series Approach”
A time series of the demand is the sequence of values of the demand (D1,D2,D3..) measured in specific time windows (periods), usually equal to each other (days, weeks, months, years)
In the Demand Components. How can you define Cyclicity?
Cyclicity: A cyclic pattern exist when data exhibit rises and falls that are not of fixed period.
E.g. cyclical products, macro-economic cycles.
In the Demand Components. How can you define TREND?
Trend: Can be increasing or decreasing and can be due to:
-Variation of the overall market volume (e.g. products in their development/death stage of their lifecycle)
- Change of the market share of a specific company (the situation of the competitors)
- change of the served market. (e.g., internationalization process)
In the Demand Components. Which causes we need to take into account for Seasonality?
The causes of seasonality is due to *Climatic,
- Costumes,
- Cyclic promotion, *Activity rate,
- extra seasonals phenomena (working days, calendar, holidays)
In the Main Methodologies of Demand Forecasting. How can you define “Time Series” Approach?
The forecast for future demand (Ft+1) is based on the analysis (possible) and extrapolation of the time series of the past demand
Ft+1 = f (Dt, .., Dt-N)
Ft+1 = f (Tt, St, Ct, et)
The function linking the forecast and the past demand depends on the specific model. There are two basic alternatives:
- models of time series extrapolation.
- models of analysis and extrapolation from time series (two steps: 1. Analysis 2. Extrapolation)
Mention some CONs related with the “Time Series” approaches
- Limited consideration of the external factors.
- Not suitable for products with no pattern or with few/no data on past depand (so, new products).
- Long set up periods might be required to select and set the models.
Mention some PROs related with the “Time Series” approaches
- Simple.
- Based on historical data (normally available).
- Easy to automate.
- If preceded by the analysis of the time series give a good level of understanding of the demand behavior.
- Easy to update the models.
In the Main Methodologies of Demand Forecasting system. How can you define Relational/Causal Approaches?
- Forecast for the future demand is based on the identification relationship between the demand and other variables influencing the demand and on the forecast or measure of these variables.
- Link can be contamporaneous or consider a time delay between cause and effect
Mention some CONs related with the Relational/Causal approaches
- Require cumbersome data analysis.
- Functional relationship (which is causal link between a group of independent variables and the demand) has to be built and validated.
- it is necessary to be able to forecast the casual factors better than the demand (dependent variable)
Mention some PROs related with the Relational/Causal approaches
- Take into account the external factors that influence or explain the demand (prince, weather condition).
- Higher intelligence on the factors explaining the demand.
- Possibility to integrate/correct the forecasts based on the time series.
In the Main Methodologies of Demand Forecasting. How can you define Qualitative Approaches
- Sales Force?
- Delphi method?
*sales force:
Marketing prepares and modifies the demand forecast on the basis of the knowledge of customer initiatives, planned promotions, macroeconomic patterns. Sales info are elaborated on the basis of a bottom-up approach
- Delphi method:
Interaction of a group of experts through a questionnaire that interactively reports the answers step by step given (anonymously), till a consensus – the most possible unanimous – is reached.
Mention some PROs related with the Qualitative approaches
- They can take into account all the factors that have never happened and can influence the demand.
- Able to create consensus involvement and “ownership” on the output of the forecasting process.
Mention some CONs related with the Qualitative approaches
- Limited capability of analysis both quantitative and formal and difficulty in managing “lots of numbers”
- High costs, meetings and external costs.
- Risk of fallacious correlations to support the thesis.
Define Forecasting error
The forecasting error at time T has been define as the difference between as the real value of the demand and the forecasted value at time T
E(t) = D(t) - F(t)
On what the accuracy Forecasting process depend?
- The aggregation level in terms of:
time, product, geographical areas (because of error compensation) - The ancticipation:
How much time before the forecast is made (the closer, the better accuracy).
Which are the demand components?
Dt: Demand value at time t Tt: Trend at time t. Ct: Cyclicity at time t St: Seasonality at time t et: noise at time t
How can you identfy the TREND?
Through the REGRESSION ANALYSIS we can identify and quantify the trend.
**BEWARE: we have to identify the kind of function y=f(t) that best suits the time series (straight line, parabola, etc.)
In order to eliminate and or reduce the seasonality and the other irregularities in the demand data, the time series should be filtered by a k period moving average (with an appropriate k)
how can you compute/identify SEASONALITY?
If we have at least a two year time series we can make an AUTO-CORRELATION ANALYSIS by computing the rk auto-correlation coefficient for the different values of “k”
The coefficient of seasonality of a generic period i is calculated as the as the ratio between the value of the demand in that period and the average value of the demand of that period
Which are the main Methodologies of Demand Forecasting.
- TIME SERIES APPROACH
- Moving average
- exponential smoothing:
Brown’s model
Winter’s model - decomposition
- arima
- RELATIONAL/CAUSAL approaches:
- Regression
- Econometrics /Input-output
- QUALITATIVE APPROCHES
- Sales Force
- Expert panels/ Delphi method
- Future scenarios/analogies
- market search, surveys, test
- marketing
Which are models of the Forecasting process organization for the following functions: FINANCE, LOGISTIC & PRODUCTION, MARGETING, SALES?
- Indipendent model
- Centralized model
- competitive model
- Collaborative model
Features of the Indipendent model for the following functions: FINANCE, LOGISTIC & PRODUCTION, MARGETING, SALES??
*Every function develops its own forecast
according to its needs
- Absolute inconsistency between forecasts
- No coordination between the functions
- No information sharing
- Low forecasting performances
