Application of "even swaps" to normalize qualitative and quantitative risk valuations
Frequently more than one risk is identified for a given work package, activity, or scenario. Depending upon the availability of historical data, the various risks may be analyzed using either qualitative or quantitative techniques. Qualitative measures are frequently expressed in ordinal terms such as “low,” “medium,” or “high.” Relative cardinal values are frequently applied as qualitative measures from a probability-impact matrix or other tool. Qualitative cardinality is used to apply a relative weight to the risk. Other risks will be assigned quantitative expected values as a result of sufficient available historical data to ascertain true probability and impact for the occurrence of an identified risk event.
Effective tools and techniques for determination of risk values have been developed for qualitative and quantitative measures. A difficulty in determining true valuation arises when both qualitative and quantitative analysis is applied to the same risks or to different risks that affect the same work package or activity. A technique to normalize qualitative and qualitative weights of differing scales must be developed. This technique is “even swaps.”
This paper applies the technique of “even swaps” to determine which risks need to be addressed. The technique is also applicable to using risk valuation to determine which projects are potentially most beneficial to the organization.
The concept of “Even Swaps” was proposed in 1998 by Hammond, Keeney, and Rafia in an article in Harvard Business Review. The application was for assistance in selecting the best of many apparently divergent choices. Even swaps uses the concept of subjective utility to trade a qualitative quantity of one attribute for a quantity of a different attribute. The result is an equalization of one or more different attributes. This attribute can then be eliminated as a determining factor in the selection of risks. A necessary step is to develop a quantitative ranking for qualitative valuations.
An example of even swaps is a project manager may be willing to trade a 20% chance of accelerating delivery of a product by one week for an increase in cost from “low” to “moderate.” The technique can be used to analyze many variables that affect a given risk, work package, activity, or scenario. The attributes that are equated can be eliminated from the decision process. The weights that are assigned to the remaining variables are used to determine the best decision.
Complex decisions are often fraught with biases that are induced by the decision-maker based upon prior experiences and learned beliefs. These biases frequently cause the decision maker to select the choice that is considered the most common or least downside exposure. Decision-makers have a tendency to either ignore or consider inaccurate, objective data that fails to meet with their expectations (Kahneman & Tversky, 1982). The application of “even swaps” helps to reduce this subjective bias.
When faced with a potential risk to a project, the project manager often is confused by the multitude of influencing factors. The risk may emanate from several sources or be of several types. The confounding effects of trying to separate the many factors and attempts to determine the most influential factors often causes the incorrect risk controls to be used to manage the specific project risk. Risks that would be actively managed on their own may be set aside with passive management in favor of other risks. The result is often improper management of risks. Another problem is that when there are too many factors, risks are often set aside due to overwhelming confusion and lack of understanding by the project manager.
The process of even swaps compares various items that may not be related. A simple example for the technique is demonstrated through determination of project selection. The only apparent commonality among the projects may that they require company assets and labor. Initially the process of dominance is used to eliminate alternatives. Even Swaps are then used to eliminate objectives.
Exhibit 1 presents a simple comparison of four projects that are competing for approval. The task is to determine which is the best choice.
The first step is to determine whether a specific alternative dominates any other. Dominance is the process of determining whether one alternative is clearly better, or dominant, with respect to another alternative. If this is the case, the lesser alternative can be eliminated from the selection. Initial observation does not yield a clear solution. The objectives are ranked across the alternatives to create a ranking table (Exhibit 2). The best value for the objective is ranked one (1). In the case of the cost objective, $700,000 is the best because it is lowest. Ties are assigned the same rank value.
A comparison of Project 1 with Project 3 clearly shows that Project 3 is preferable to Project 1 due to the two rank 4 in Project 1 while Project 2 has no rank 4 and one rank 1. This is known as practical dominance because the result is obvious. The other three projects are not as clear. We now apply even swaps to the resulting alternatives. Hammond et al. state that, “If every alternative for a given objective is rated equally…you can ignore that objective in making your decision.” We now replace the actual values for the remaining alternatives because there is no longer a practical dominance among the ranked alternatives.
In order to draw comparisons, we will equalize some of the objectives. The project manager may determine that for Project 2 a cost increase of $200,000 is sufficient to reduce the project duration by two months. The project manager is willing to spend an additional $300,000 in order to engineer an improved cash flow to increase the NPV by $2,500. The result is that the Cost objective is no longer a factor.
The project manager next determines that increasing the for Project 4 increasing Business enhancement to High is worth a reduction of NPV by $1,000.
The project manager next determines that increasing the for Project 2 increasing NPV by $1,000 is worth a reduction of adding two months to the duration.
The clear alternative is to select Project 4 based upon the eight-month duration.
Application to Risk Management
In order to demonstrate the application of even swaps to project risk management, let us assume that we want to explore the risks involved in a large purchase of personal computers by a company. For example purposes we must identify potential risk types and risk sources.
Risk identification is not included in this paper. An assumption is made that risks have already been identified. The processes of dominance and even swaps are used to determine which risks are most important and which controls are best used for the risk response.
In the example of procurement of personal computers for the organization we may identify certain risks. A necessity exists to determine which risks are most important with respect to the objectives of the project and overall corporate governance. Exhibit 7 summarizes some of the risks that have been identified. The subjective valuations have been obtained from the sample probability-impact matrix. Quantitative values are expressed in terms of expected values.
Risks can be examined from multiple dimensions. Even swaps can be used to determine the risks for which control must first be developed. In the examples of the procurement of personal computers, we may develop the following risk list.
We assume that the identified risks are (1) insufficient memory is specified, (2) the display screens are of too low grade for the applications, (3) insufficient quantity are specified, and (4) the incorrect software was loaded onto the computers. The objectives are: (1) the purchase is within budget, (2) delivery is on schedule, (3) no product is inventoried in a warehouse, and (4) the computers have sufficient performance longevity, and there is uniformity across the organization. For purposes of this paper details of how the analysis was performed to arrive at the valuations are not shown. It is assumed that quantitative values are based upon historical information.
The next step is to create a ranking matrix in order to search for dominance. For risks the alternative with the greatest value is assigned the lowest rank and the alternative with the lowest value is assigned the highest rank. The intent is to eliminate the risks with the least values. In this manner, greater adversity will dominate lesser adversity.
Since Improper software practically dominates Insufficient Quantity, Insufficient quantity can be eliminated. Likewise, Improper software practically dominates Low-grade screens. Therefore, low-grade screens can be eliminated as the most important risk to procurement of personal computers.
Now that the risks have been eliminated, the next step is to redraw the original table excluding the eliminated risks. The even swaps are now applied to the result in order to normalize the objectives.
In this example, let us assume that we are willing to delay the schedule by one day in order to reduce inventory costs by $400. When this is applied to the table, we have the following result. Schedule is no longer a factor.
A decision is made that a sacrifice on budget to a factor of .80 is acceptable for insufficient memory to increase uniformity very low risk with high impact, a weight of .08.
At this point, two solution alternatives are available, a new ranking table can be drawn to determine whether one alternative dominates the other, or even swaps can continue. A new ranking table yields the following in Exhibit 14.
Clearly, Improper software dominates Insufficient memory. As a result, a decision is made to focus on improper software loads as the greatest adversity to the procurement of new personal computers.
If even swaps had been applied, a tradeoff to increase the longevity costs by $3,000 in exchange for reducing the uniformity exposure to .14.
A subsequent trade may be made to decrease inventory costs by $100 in exchange for increasing longevity costs by $1,000.
The result is that Improper software loads are the first risk that must be examined and controlled.
The process of even swaps is an excellent tool to normalize qualitative and quantitative values that are assigned to a single risk. It is applicable to determining which risks are most important to the project. Using the technique of even swaps forces the risk analyst to view risks in terms of similar objectives to the project and overall corporate governance. The identical technique can be used to isolate the major sources of project risk.
Hammond, J. S., Keeney, R. L., & Raiffa, H. 1998. Even Swaps: Rational Method for Making Trade-Offs. Harvard Business Review 76 (March-April), pp. 137–150.
Kahneman, Daniel, & Amos Tversky. 1982. The Psychology of Preferences. Scientific American (January), pp. 160–166.
Proceedings of the Project Management Institute Annual Seminars & Symposium
October 3–10, 2002 • San Antonio, Texas, USA