Academic journal article The Journal of Business Forecasting Methods & Systems

The Storage of Forecasts

Academic journal article The Journal of Business Forecasting Methods & Systems

The Storage of Forecasts

Article excerpt

Designing a reporting system should be based on the needs of users ... store data at the lowest possible levels of granularity for the greatest flexibility .... between the time when the Base Statistical Forecast is released and the time when Consensus is reached, there may be upwards of 10-20 different forecasts.

he database design and storage requirements of forecasts are important considerations in building efficient forecasting systems. This article deals with these issues.

DIMENSIONS OF FORECASTING

To design the structure of a database for saving forecasts, one needs to understand the context in which the forecast was created and used. To illustrate the importance of context, consider the following question: Question: What is the forecast? Answer: 10,000 Units

The above answer doesn't tell anyone much because the number has not been placed in any kind of context. A number alone provides no clue about Who, What, When, or Where.

Each company views forecasting in a unique manner because ofthe need to match forecasts with internal culture and processes. While the specifics may vary, the organization of forecasting data can be reduced to a few common themes. All forecasts have some measure of time associated with them by definition. At an aggregate level, all companies also have at least one product or service and operate within some geographical boundary. Finally, all companies have an audience for their goods or services. Each of these four components, Time, Product, Geography, and Customer, represent dimensions upon which forecasting data are built. They correspond to four basic questions regarding forecasts:

When is the product shipment or service going to occur? (Time)

What product is being shipped or delivered? (Product)

Where is the product going? (Geography)

Who is receiving the product or service? (Customer)

Answering these questions provides the context of a forecast. The answers can Mr. Power is the Director of System Integration at SHL Systemhouse. In this capacity, he helps clients to define and install management support systems. Prior to joining SHL, he worked for a software products company and held several analytical and research positions in the consumer packaged goods industry. He has extensive experience with syndicated data as well as with internal shipments and financial information. range from very broad definitions to very specific instances. For example, at one extreme there is the statement of "How much are we going to ship next year?" At the other extreme, forecasts may be made at very low levels of granularity, e.g. electric utility demand by hour for rural households. Implicit in both of these statements is a definition regarding Time, Product, Geography, and Customer. In the first case, Time is 'Year' and the remaining three dimensions are `Total Company,' i.e. all products to all customers in all regions. In the second example, much more detailed definitions have been made: Hour of day, kilo-watt hours, rural geography, and domestic customers.

The first design decision is to determine the dimensions by which to store and manipulate the forecast. If one uses basic questions like those posed above, then there are probably three to five fundamental dimensions for most forecasting systems. Each dimension can then be assigned levels or attributes that refine the definition of Time, Product, Geography, and Customer as discussed below.

Two other early decisions in designing forecasting systems are: at what level are forecasts made, and at what levels are forecasts reported. The answers are by no means necessarily the same. The decision criteria for creating forecasts often represents a balance between technical (statistical) efficiency and business effectiveness. The problems associated with forecasting with data that are too erratic because of sparsity are well known (e.g. dealing with time series that are mostly zeroes) as are those of forecasting at too high a level (e. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.