Education has recorded a $12.9 billion liability for loan default losses and interest subsidies under its Federal Family Education Loan (FFEL) Program. Estimating and recording this liability is not only a critical step for preparing financial statements, but it should also be the outcome of a credit management process through which the Department assures individual loans are administered properly and cost effectively. The accuracy of the liability and effectiveness of the Department's oversight of the FFEL Program's credit management process ultimately depend upon the veracity of underlying data on individual guaranteed student loans. For example, historical data about the characteristics of the borrowers who defaulted, about how much of a claim payment was made and how much was subsequently recovered through collection efforts, are crucial to understanding why defaults occur and how to maximize recoveries. Analysis and use of this data for management purposes is particularly important when, as with the Department's FFEL Program, the administration of credit and collection of defaulted loans is in the hands of a variety of third-party intermediaries. We noted, however, that despite recent efforts to clean-up FFEL Program loan data, there are still many instances where key historical data is either missing or where questionable data cannot be explained. Further, these data problems could also affect the allowance for loss on direct loans, since historical data on guaranteed loans is the primary basis for calculating the direct loan allowance.
As shown in Figure 1, many FFEL Program operations are performed by Lending Institutions and Guaranty Agencies (GAs). Key data necessary to calculate Education's liability estimate flows from approximately 8,000 lending institutions to 41 GAs and then to Education. Since inception of the FFEL Program audits for fiscal year 1992, the GAO and OIG have reported that based on their testing of loan documentation, significant errors existed in the data transmitted to Education such that Education cannot be assured that its liability estimate is materially correct. Our testing also revealed errors or missing data in the database used to calculate the liability.
Financial services institutions subject to Federal Regulatory oversight are required to maintain accurate loan and accounting data and have a process in place to analyze and evaluate such data to ensure proper credit management techniques are applied.
The absence of complete and accurate data causes regulators and auditors to question not only whether the institution is being operated in a "safe and sound" manner, but also the accuracy of related loss reserves. Under generally accepted accounting principles, data underlying loss reserves must be reliable and sufficient to explain; among other things:
Education recently engaged a contractor to develop a model for estimating the liability for FFEL Program loan guarantees and to develop the estimate as of September 30, 1995 for inclusion in Education's fiscal year 1995 financial statements. Data used by the contractor in calculating Education's FFEL Program liability estimate for fiscal year 1995 were derived from the sources as shown in Exhibit 2:
|Key Data Components||Source||Data Availability|
|Defaults||NSLDS1||FY 1990 to present|
|Collections||NSLDS||FY 1990 to present|
|Outstanding Balance||NSLDS||January 1996 to present|
|Net Guarantees||Databook||FY 1965 to present|
1 National Student Loan Data System
With the exception of net guarantees, only limited historical data was available to estimate the liability. Having accurate historical data about when and in what amount common groups (or cohorts) of loans default, repay or receive subsidies is important to understanding and managing credit risk as well as for calculating accurate loss reserves for financial statement purposes. Such information is also essential for assuring that credit reform subsidy calculations are as accurate as possible. Because of the data constraints, the contractor, in developing the fiscal year 1995 liability estimate, applied two general assumptions: 1) that available data can be used as a basis for constructing a historical performance pattern; for example, collection data from 1990 and forward can be used to construct prior year collection patterns; and 2) that any data errors, either previously identified or going forward, will essentially "net out" (i.e., they are unbiased) and thus will not have a material effect on the liability estimate.
Despite the efforts of the contractor, we do not believe the uncertainties created by the lack of and reliability of data have been overcome. In particular, there is no pre-1990 information available in the database. Education is still in the process of fully populating NSLDS with 1990 and later loan transaction data. Guaranty agencies were not required to report collections on defaulted loans for the entire six-year period from 1990 to 1995. However, some collection information was reported to the NSLDS at the guaranty agencies discretion. This means that Education had only six full years of actual collection data to use as a basis for projecting collections over the life of the loan. However, the majority of collections occur after year six of the loan. As such, there was very little data on which to assess historical collection patterns, and the contractor had to estimate the relationship between collections and defaults. This estimate was primarily based upon estimated defaults and estimated collections rather than actual defaults and actual collection data. Because collections are such a critical and material part of the liability calculation, this is simply not enough data to conclude that the liability could not vary materially.
With respect to the potential effect of data errors, in developing the FY 1995 model, the contractor reviewed the GAO and OIG identified errors from FY 1992 (a 34% error rate) and concluded that when netted together, the errors did not cause a "bias" in the assumptions underlying the liability estimate and thus did not have a material dollar impact on the overall liability. Implicit in this assumption is that the type of errors discovered by GAO and OIG and the data fields they affect would hold true for subsequent years as well. But there was no testing performed on post 1992 data to determine if this was in fact the case. The use of this assumption without corroborating information creates uncertainties about the liability estimate. This is particularly true when the potential effect of these errors is considered along with the uncertainties introduced by the lack of other important data.
Education's approach for the FFEL liability estimate for fiscal year 1995 also did not include adequate analysis of the underlying variables that explained trends, such as changes in default behavior or collection patterns due to the limitations of the available data. We were concerned that trends that existed in data produced by the model could not be explained, other than anecdotally. For example, the model shows that default rates have dropped significantly for all school types. These trends could be due to an improved program, increased oversight of institutions, changes in the economy, unreliable data, or some other cause. We were not provided with empirical support that explained these trends in the data.