How does the DuPont Model—a hybrid of which provides the methodology behind the Fortnightly 40 rankings—actually work? The author shares lessons learned during implementation of the hybrid model this year.
Jean Reaves Rollins is managing partner of The C Three Group. Contact her at email@example.com.
The DuPont Model, like many financial analysis techniques, goes in and out of style. However, many analysts keep it in their bag of tools for evaluating the fundamentals of asset-heavy companies. It has a proven track record, is relatively simple, and provides intra-industry apples-to-apples comparative results. It often is used in conjunction with total shareholder return and other analyses to triangulate corporate values or performance.
During our analysis of the Fortnightly 40, the correlation between the results of the DuPont Model and stock-market results for the same group of 102 companies was questioned. This called for a closer examination.
Using the base group of investor-owned electric, gas, and pipeline utilities, we projected a value of $100 invested in each as of Jan. 1, 2003, and what the value of that investment would be on Dec. 31, 2005 (adjusted for splits and dividends)—the same time frame as this year’s Fortnightly 40 analysis.
Table 1 presents the top 10 in stock-price appreciation.
Using a simple regression analysis, we correlated the ranking results for all 102 companies of the equity analysis against the results of the modified DuPont Model. We found essentially no correlation—or 0.028, to be exact.
However, nasty outliers must be dealt with. We looked at the top 10 again, and decided that the top three performers were seriously struggling with the post-Enron market freefall. All three were trading at or near their historic lows. So we tossed them out of the analysis. When we did this, suddenly, DuPont and stock performance became more highly correlated, at 0.302.
For balance, we then tossed out the bottom three performers (those companies with negative shareholder value, or those that were in bankruptcy during any of the three fiscal years under analysis, had been eliminated from the beginning). Eliminating the bottom three had minimal impact on the correlation, going from 0.302 to 0.29.
So what does this mean? Working with a “clean” data set of companies shows that their DuPont scores relate to the stock values of these companies, and that the higher the Dupont score, the larger the increase in equity values during the same period.
However, how did we choose the outliers to come up with a clean data set? Fig. 1 shows a composite index of the original 102 companies in our sample. It clearly shows that few companies had escaped the decline in the overall market and in the U.S. energy utility market, in particular, during the 2002-2003 timeframe.
Removing all of the companies that had taken significant equity price hits during this period would have left a sample too small for analysis. We removed the top three, believing that their equity price gains were so above average it was likely to skew the results of the analysis. Generally, we would suggest a much more analytic approach to determining what are truly outliers.
We also looked at a number of other potential correlation sets. Table 2 presents the results of this analysis for: a) the complete set of companies; b) the set with the companies with the top three equity increases removed; and c) the set with the top three and the bottom three equity price increases removed.
We learned several lessons through this process:
1. The DuPont Model is highly consistent in its results and the relationship with variables used in the formula itself or with totally independent variables. It also is highly consistent between sample sets. Manipulating the sample set did not have a major impact on results.
2. Equity price correlations are much less consistent, and the correlation results easily can be manipulated by the sample set chosen. Thus, it is much more difficult accurately to interpret true relationships between changes in stock prices and underlying corporate financial metrics.
3. The negative relationship between dividend policies and stock price increases was one area where manipulating the data set did not radically alter the results. However, the old adage, “correlation does not imply causation,” should be remembered when interpreting these results.
4. “Bottom fishing” can be very rewarding financially. Virtually all of the top 10 companies in equity price appreciation were trading at or near their historic lows on Jan. 1, 2003, irrespective of what the company’s DuPont rank may have been. So while the DuPont model is an excellent tool for providing information on how well a company may be managed, at least from a historic perspective, it may not be the best predictor of how a sometimes irrational stock market may value a company or industry.