SUPPLY CHAIN MANAGEMENT – MODELING CONTRACT RISK
Contract risks exist in nearly all economic sectors; however, some contracting and procurement organizations do not generate plans or perform rigorous evaluations on their suppliers to minimize this risk. One of the main goals in managing contracts is to avoid interruptions in the services performed by suppliers. To meet this goal, supply management professionals must evaluate each supplier that may have a contracting agreement with their organization.
From a technical point of view, there are several elements — certifications, tangible evidence, site visits and the like — that help supply management professionals minimize a potential breach of contract. This is not the case when we look at the financial side.
While many organizations include bidder’s financial statements and financial ratio evaluations as part of the technical qualification in a sourcing process, there is no standard methodology to quantify or evaluate the bankruptcy or insolvency risk of those bidders. Also, depending on the length of the contract, the financial stability of suppliers may have changed over time. Discriminant analysis may be the answer for those looking for a solution to this problem.
DISCRIMINANT ANALYSIS — WHAT IS IT?
Discriminant analysis is the most common statistical method used to predict bankruptcy risk. It is a statistical model used to classify an element into two or more predefined groups. For this article, the predefined groups are those with high risk of insolvency and groups with minimal risk of insolvency. This method is used primarily in analyses where the dependent variable is presented in a qualitative form. To have a consistent outcome, the model needs to be applied on a sector-by-sector basis to achieve an apples-to-apples comparison.
BUILDING THE MODEL
Building a discriminant analysis model is a five-step process involving: 1) identifying the data, 2) preparing the data, 3) generating the function, 4) understanding the results and 5) choosing the predictive value. To illustrate the steps, consider a contracting company that expects to initiate a variety of sourcing processes related to trucking services. The company needs a tool to qualify the financial situation and indicate the likelihood of bankruptcy for each supplier it will eventually invite to participate. The outcome will be referred to as a dependent variable (referred to as “S”). The company also wants to calculate this likelihood based on specific financial ratios for each supplier — these ratios then become the independent variables.
STEP 1: INFORMATION REQUIRED FOR THE ANALYSIS. To build the model, the company needs two main pieces of information:
- Two groups of companies: one group that actually went into bankruptcy and another group that didn’t during a specific period of time for the same sector being analyzed (in this example, trucking services). This is a key requirement. Without it, the formula cannot be built.
- Financial statements for the past two to five years for the suppliers identified in the previous point. You can gather this information from public sources if the companies are publicly traded, or request the information from the suppliers. For example, if the supplier is privately held, you may need to request specific data; or, if the supplier is publicly traded, you can gather the information by examining the supplier’s filings with the government and other sources of publicly available information.
STEP 2: PREPARE THE DATA FOR ANALYSIS. The next step is to perform calculations using selected financial ratios, which allows us to generate the model, as shown in Figure 1 below. The number and type of financial ratios may vary by sector, but the common rule is to include as many as possible. In our example, we identified 10 financial ratios (see the sidebar on the right for their definitions) and separated the suppliers into two groups of three companies each. Then, we used the past financial statements, gathered in the previous step, to calculate each of the financial ratios for each supplier in the two groups.
STEP 3: GENERATE THE DISCRIMINANT FUNCTION. Once the financial ratio calculations are complete, a discriminant analysis can be conducted using statistical software currently available on the market (for instance, SPSS, SAS or Statistics). The software will create a formula (the discriminant function) that separates each potential supplier in both groups based on the independent variables (in this case, the financial ratios). If successful, the software identifies the discriminant function as well as a series of coefficients that measure how well the function discriminates or categorizes the groups.
STEP 4: UNDERSTAND THE RESULTS. After entering the data into the software, the program provides a variety of information, including:
- Discriminant function. This formula separates a concept (in this case, suppliers) and classifies them into groups. The function reveals coefficients (in this case, the financial ratios) and a constant (an unchanged value).
- Wilk’s lambda test statistic. This statistic shows how separated the groups are on a scale from 0 to 1. If the value is closer to 1, it will indicate low discrimination. On the other hand, if the value is closer to 0, there is a high difference between the groups. The implication of a low discrimination is that the groups cannot be easily compared and contrasted. Thus, a different financial analysis tool should be used.
- Group centroids. Centroids represent the average value of the scores for a specific group. If centroids are quite different, then the groups have a high discriminatory power and vice versa. Again, a low discriminatory power reduces the comparison effectiveness of the discriminant analysis.
Going back to our example, the contracting company used the information on the table in Figure 1 and entered the data into a statistical software application. The software provided the discriminant: S = (2.15 x S3) – (0.12 x S6) + (1.20 x S7) + (114.5 x S8) – 0.14.
Within the discriminant function, the dependent variable S equals the score for the “n” supplier. The software identified coefficients including 2.15, 0.12, 1.20 and 114.5. And the software identified the independent variables to be used in the discriminant function as S3 (acid test ratio), S6 (working capital turnover), S7 (operating cycle) and S8 (debt ratio). From the 10 ratios analyzed, the software determined that only these four ratios were needed to
Discriminant Analysis Model
identify in which group the specific supplier is classified (high risk of insolvency or minimal risk of insolvency). In addition to the financial ratios, the function includes a parameter — the constant, which equals 0.14. However, it is the coefficients that weigh the importance of each ratio.
Our example also indicates a Wilks’ lambda test statistic (∧) that equals 0.002. Because the number is much closer to 0, this means there’s a large difference between the suppliers classified as solvent and insolvent.
The group centroids identified in our example were calculated as 37.24 for the onbankruptcy group and 1.55 for the Bankruptcy group. The difference in values between the two groups means they are highly discriminatory (there’s a large difference between the numbers to gain a clear comparison between the two groups). These two numbers are also critical in the next step when choosing a predictive value.
If the formula/function has a low discrimination power (for example, a Wilk’s lambda test statistic that is close to 1 or centroids that are close in range to one another), then it won’t work correctly, and it couldn’t be used to separate the suppliers into the groups. In some cases, this may occur and, in such cases, the discriminant analysis is not the appropriate tool.
STEP 5: CHOOSE THE PREDICTIVE VALUE. If the software provides the expected results — meaning it provides a function that can efficiently differentiate your predefined groups — the next and final step is calculating the predictive value. This value is used to compare the different scores the discriminant function provides for each of the suppliers under analysis. Depending on where the specific score falls, that particular supplier will be classified into one of the two groups.
Identify this predictive value by calculating an equidistant point between the centroids: Nonbankruptcy centroid (37.24) + Bankruptcy centroid (1.55) ÷ 2 = 19.39. Therefore, if the score S for a supplier is below 19.39, this supplier will be classified within the Bankruptcy risk group. On the other hand, scores above the predictive value are categorized within the Nonbankruptcy group. The function and the predictive value are valid only for the trucking services market and should be updated with new historical data periodically to guarantee accuracy in the categorization.
The supply management professional or category manager may use this valuable information in a pre- or post-award contract phase.
Pre-award phase. From a pre-award contract perspective, supply management professionals can use the bidder scores as part of the evaluation analysis in a sourcing process (that is, assign weights depending on the value each bidder obtained). In our example, the contracting company that built the model is now in the middle of a tender process for trucking services in the southern region. As part of a qualification process, “financially healthy” was included as a factor in the evaluation matrix, and scores need to be assigned to four suppliers. In our example, a scale of 50 (scores below 19.39) and 100 (scores above 19.39) representing poor and high financial health was used. Financial statements were requested from each one of these suppliers as part of the tender package.
Financial ratios S3 (acid test ratio), S6 (working capital turnover), S7 (operating cycle) and S8 (debt ratio) were calculated based on the financial statements and introduced in the function: S = (2.15 x S3) – (0.12 x S6) + (1.20 x S7) + (114.5 x S8) – 0.14.
Pre Award Phase
Results are shown in Figure 2 above.
Consequently, bidders A and C received scores below the predictive value of 19.39 and were considered suppliers with a high risk of bankruptcy. Therefore, they are given 50 points in the qualification matrix. On the other side, bidders B and D are awarded with 100 points due to their financial health (their S scores are above the predictive value of 19.39).
Post-award phase. In a post-award contract situation, the same philosophy can be used as a monitoring system for current suppliers with contracts in place. This helps the contracting organization anticipate risk situations with a particular supplier and take preventive actions to avoid or minimize a potential service disruption. In the example in Figure 3, below, the same contracting company decided to award the trucking services contract in the southern region to supplier D. Two years have passed since the award, and the supply management organization has been recalculating the S score for this supplier on a year-by-year basis using updated data for each year to monitor its financial stability.
Post Award Phase
The downward trend in the supplier’s S score, as noted in Figure 3, illustrates that something occurred with this supplier since signing the contract two years prior. This clearly indicates that there is a high risk of bankruptcy that might affect the contracting organization. Supply management professionals can use this information to mitigate risk by trying to understand the current situation of supplier D and help them get back to the right path, or identifying a backup supplier in the event supplier D is not able to deliver the services requested in the future.
Supplier insolvency or bankruptcy risks remain prevalent in the current business environment. No matter how it is used, discriminant analysis is a tool that provides consistent and accurate information on an organization’s financial health at any point in the procurement process, providing supply management professionals with an additional and statistically correct option to evaluate or monitor their suppliers. Furthermore, the tool will improve the overall understanding of an organizational supplier base from a financial point of view, making the sourcing decisions more robust.
Going forward, the same analysis may be applied to other areas of the supply management function where predictive modeling for categorization is needed. A good example is on-time delivery suppliers versus suppliers with recurrent delivery delays. If we could generate a mathematical function that allows us to identify if a supplier will deliver on time, we could use this logic as a sourcing award decision based on the required on-site date for a specific material — thus optimizing the procurement process and guaranteeing material availability. There are many areas waiting to be analyzed. The possibilities are endless, and the tool is already here.
Always Yours——–As Usual—— Saurabh Singh
Source: August 2010, Inside Supply Management® Vol. 21, No. 8