1. Filtering Process: You pick a certain criteria and set a “must have” level. My car must be red or blue. My cereal must have 0 trans fat. No matter what decision you are making, all this process does is narrow the field. When you get to the end of this filtering process, you only see the candidates that have met all of your must-have criteria and they are not in any particular preference order. This is any easy way to narrow the field of candidates, but not a very good way to make decisions.
2. Modeling Process: The second method uses an importance ranking and factor model process. That may not be what you call it but that is what is going on inside your head. This method produces much better information, but is harder to implement. Back to our car example: We might prefer a blue car over a red car and a V-8 over a 6 cylinder engine. So obviously a blue car with a V-8 is preferable to a red car with a 6 cylinder. But what about a red V-8 or a blue 6 cylinder? Suppose blue paint costs an extra $1000 and a V-8 costs an extra $5000. You may find that you prefer a red car with a V-8 that costs $41,000 over a blue car with a V-8 that costs $42,000. Someone else may make the opposite choice because the color is more important than the extra cost. To make the decision, you need a process that can simultaneously evaluate multiple criteria based on how important each one is to you. Ideally, the final result is a ranking of each distinct product based on importance and value of each criteria combined into a single score - i.e. a factor model and importance ranking.

Many investment decisions are based on the Filtering Process. The major reason is that filtering or screening (both names are used) has been in widespread use since the late 1980s when PCs became common. Software that uses this method was developed and it became ingrained in the financial services industry. For proof of how prevalent the screening process has become, just visit the web sites of advisors showing how they select managers and mutual funds.

As shown above, the Modeling Process is superior to the Filtering Process, so why isn’t it used more frequently for making investment decisions? In our simple example we only had three variables for our car: color, engine, and price for the options. And we only had two choices for each. Now think about a complex decision such as evaluating mutual funds where there are hundreds of different variables and thousands of choices and combinations. The complexity is daunting, even if there are significant resources to devote to building factor models.

In evaluating mutual funds, there are thousands to compare and the data changes at least monthly. Using the Filtering Process, you end up evaluating only a very small subset of the available funds. In fact, because there is no overall ranking, you are actually forced to reduce the total to a small number because the final analysis process is so labor intensive. This is illustrated in the chart using four variables: Expenses, Return, Sharpe Ratio, and Worst 4 Quarters.

In this example, expenses must be less than 1.25%. A fund that has expenses of 1.26% and excellent returns, Sharpe ratio, and worst four quarters will never survive the first cut. But it may be one of the best candidates when considering all four attributes. The saying often used about screening is “We know we may miss some of the best funds. We aim to eliminate the losers and only some of the real winners.” 

If we adapt the Modeling Process for investment decisions, that sacrifice is no longer necessary. But as stated above, that can be a difficult task. There are two major issues that must be resolved when building factor models:

1. Scoring System: A relative scoring system must be developed that allows the advisor to compare criteria that are expressed in different units. Sharpe is a ratio while the other three are percentages and each percentage measures something different. A scoring system must provide a way to compare these by using a relative score for each variable. 
2. Factor Weighting: Secondly, the relative importance of each variable must be determined. Intuitively you know that a conservative investor will place more importance on risk and an aggressive investor will put more importance on return, but it can be difficult to determine exactly how much more or less. Not only must the weight of each criterion be considered, the relationship between the variables must also be considered as illustrated in the table below.

In reality, each individual weight has an impact on all of the relative weights. This works like a correlation table in an asset allocation study. In that situation, no one would think of changing one variable and expect it to have no impact on the relationship of the other variables. But if you simply weight the criteria in a factor model, you have really done just that.

As you can see, complex multi-factor models are virtually impossible to “build” in your head because the human brain cannot process that many variable relationships simultaneously. Some people build models using Excel and either set the weights equally for simplicity or assign values as they deem appropriate, but they rarely consider the correlation between the factors. Building complex scoring systems and correlation tables for weighting factors, then updating them regularly, is simply not cost effective for most investment advisors and analysts.

To deal with the human limits and time constraints, Klein’s K⁴ Fund Selection employs a patented derivative of adaptive conjoint analysis so that the user never has to evaluate more than two variables at a time. The software then recombines the “two at a time” answers into an overall relative importance score for each criterion, eliminating the need for manual construction of correlation tables in spreadsheets and providing an automatic scoring system that ranks the funds.

With Klein’s unique K⁴ decision process, an advisor or analyst can easily evaluate thousands of funds on multiple attributes. K⁴ ranks all the funds simultaneously in a single, comparative list and then the user can filter the results for “must have” limits. The filters are applied at the end to show what funds were eliminated by the filters and why.

The Filtering Process is a linear independent process. Each screen is applied sequentially and independently of every other screen. A similar process is used for most investor profiling for the purpose of determining an appropriate portfolio or strategy. An investor is asked a series of independent questions and each answer is worth a certain number of points. The points are added up and cross-referenced with a list of investment options to determine the correct portfolio or strategy for the investor. Just as in the Filtering Process for investment product decisions, there are flaws with this linear independent process:

1. There is no way to assign differing levels of importance to the questions for different investors. Each investor gets the same points for the same answer as any other investor.
2. By its nature, this process averages conflicting answers. For example, an investor who answers one question aggressively and another question conservatively will have the same average score as an investor who answered both questions with a moderate response. This means that the first investor has an unidentified conflict that may not be recognized and considered in the recommendation.

The conjoint analysis approach mentioned earlier has been used for decades in the field of market research. The core process asks consumers about their preferences for different attributes. In our earlier car example, the trade-off question might be the consumer’s preference for the desired color or the desired engine. If we translate this concept into an investor profiling question that allows the investor to differentiate the degree of preference, we have a trade-off question structured like this:

In this case the investor is asked his or her preference for achieving the desired level of investment return or desired level of investment income. In the second question the trade off is between achieving the desired level of investment income and not exceeding the maximum desired chance of loss. From these combined answers, a relationship for this investor can be determined between investment return and chance of loss. We now have a process that links the investor preferences in a nonlinear and dependent manner.

The result yields a unique importance level for each attribute based on the individual investor’s responses. An example of how this can be expressed in an easy-to-understand presentation is shown below.

The return is the single most important attribute, but the combination of the importance level for the two risk attributes shows this investor is also very concerned about risk. This result can also be carried a step further: Each portfolio has a certain performance level for each attribute. Just like the investment product decision, we can now take the investor importance levels and rank the portfolios in a preferred order for each individual investor.

With this process, consistency of answers can be calculated and conflicts resolved. The end result demonstrates how well the portfolio meets the acceptable level for each attribute and shows how well the portfolio “fits” when all of the attributes are considered together.

Making sure an investor is matched to the most appropriate portfolio and evaluating investment products are vital steps in the overall investment decision process. It is time to significantly improve these decisions with a change in thinking from a linear independent evaluation process to a nonlinear dependent factor evaluation process. Although this may be a new way of thinking, new thinking improves our analysis and advances our industry. A fairly recent example of this is returns-based style analysis first introduced in 1992 in a paper by Dr. William F. Sharpe. The quadratic optimization used for RBSA had been used in operations research for decades, but Dr. Sharpe was the first to use this technique for this particular application. Now it is widely accepted as a major component for evaluation of managers and mutual funds. In the same way, adaptive conjoint analysis has been widely used in market research for decades and the time has come to capture the power of this technique to make better investment decisions.