個案總覽

This technical note introduces (1) approaches to forecasting in general, (2) simple moving averages and exponential smoothing, (3) accounting for seasonality in forecasting, (4) accounting for trend in forecasting, and (5) implementing a forecasting model. Holt and Winter models for exponential smoothing are included.

A division president of a Fortune 50 company must decide whether to initiate an immense capacity expansion. The critical question is whether revenue from the company's major product line will continue to grow rapidly. An opportunity is available to conduct some market research prior to making the decision. This case introduces the steps of decision analysis: decision diagramming, assessing monetary consequences, assigning probabilities, and evaluating expected monetary value to "fold back" the decision tree. The case also develops the ideas of value of information (clairvoyance), control (wizardry), and sensitivity analysis (breakeven probability).

B Case to UV0725

The manager of a large downtown hotel has to decide whether to accept 60 additional reservations or not. If she accepts, she will be overbooked and face certain costs if all the people holding reservations show up. The manager must forecast, based on historical data, how many of the people holding reservations will show up and then decide, after taking into account the cost involved, whether to take the additional bookings. The case can be used in a class on seasonality and exponential smoothing in time-series forecasting.

The president of a well-established manufacturer of mattresses has asked his assistant to recommend whether to consolidate plants of three different locations. He must decide not only which criteria are most useful in making such a decision, but also how to weigh the different criteria in coming up with a single decision.

This technical note considers how one makes choices in situations where there are multiple objectives and attributes that measure performance against these objectives. In some instances, given attributes may be important because they measure impacts of a choice on specific stakeholders. The note considers elimination of alternatives by aspects, dominance, and some decision rules that do not require compensatory tradeoffs (lexicographic, satisficing). Methods for "rate and weight" (linear additive scoring rules) are described and illustrated.

This note teaches the student how to account for seasonality in time-series data. All the necessary steps to (1) deseasonalize, (2) forecast with deseasonalized data, and then (3) re-seasonalize the forecast are illustrated with examples from the coal industry.

The president of a large and established manufacturer of laboratory equipment has to decide whether to invest $1 million for 30% equity in a start-up company in the field of lab robotics. The agreement would also allow his company the right to market the product. He already has a spreadsheet that projects the best guess of the future scenario and calculates several measures of performance (ROS, ROE, ROI, NPV, and IRR). He must decide which of the criteria are most useful. A relevant-cost issue that is introduced must be resolved, because it makes a big difference in the NPV. In the supplement, some background material is provided for a forecasting/judgmental assessment exercise based on this decision. The supplement could, assuming students have already been introduced to this topic, form the basis for a short workshop (an hour or less) on judgmental probability, or it could be used with a note on cumulative probability distributions for an introductory class on the topic. (The B case number is UVA-QA-0383, and a supplement to the A case is UVA-QA-0384.)

Four years after the A and B cases (UV0528 and UV0530), the annual plan has drifted greatly from that suggested by the analysis in the B case. How can the decision-analysis methodology be simplified to provide routine planning guidance?

With the assistance of a temporary graduate student, the director of corporate research uses multiattribute decision analysis to find a plan with greatly improved projected performance (see UV0528). The methodology for eliciting probabilities, assessing trade-offs and risk preferences, and finding the most desirable plan is detailed. (A follow-up case, which takes place four years later, is UV0531.)

In this B case, the initial results from a Monte Carlo simulation seem to conflict with the critical fractile result discussed earlier in the case. The manager must figure out why and then make a decision.

The president of a natural-resource exploration company has to decide whether to invest in a new drilling opportunity. He already has a spreadsheet that projects the most likely scenario for the well and calculates the NPV and internal rate of return. However, six uncertainties are discussed by the president and another potential investor. He also has prepared a spreadsheet for a couple of downside scenarios--one where gas is not able to be produced after the well is drilled and a second where gas is produced but all other uncertainties are at their 1% worst possible values. A student worksheet file is available for use with this case.

This case considers the EPA's bubble policy, which allows pollution to be managed on a plantwide basis rather than stack by stack. A pollution constraint is added to a linear-programming model that may be used to plan production and the control of pollution.