volume 57 article #4

Stabilizing and Extending Qualitative and Quantitative Indicators of Creditworthiness in Agricultural Credit Scoring Models

Michael P. Novak and Eddy L. LaDue

Michael P. Novak is manager of agricultural finance with the Federal Agricultural Mortgage Corporation, Washington, DC; Eddy L. LaDue is a professor of agricultural finance, Department of Agricultural, Resource, and Managerial Economics, Cornell University.

Abstract

This study introduces quantitative and extended indicators of creditworthiness for use in credit scoring models. Annual, two-year average, and three-year average indicators of qualitative creditworthiness, as well as annual, two-year average, and three-year average indicators of quantitative creditworthiness, are calculated using the coverage ratio. The respective indicators of creditworthiness are incorporated into annual, two-year average, and three-year average credit scoring models. The results indicate that the average models’ creditworthiness prediction rates are more accurate than the annual model’s creditworthiness prediction rates. The results also suggest that the projected coverage ratio may be a viable, alternative, quantitative indicator of creditworthiness.

Key words: extended creditworthiness, quantitative creditworthiness, qualitative creditworthiness, multiperiod model, credit evaluation, coverage ratio.

Article <top>

An empirical concern about credit scoring models is how to define and measure creditworthiness. Typically, agricultural credit scoring models replicate lenders’ subjective classifications or estimate default versus nondefault as accurately as possible. In order to overcome the inherent subjectivity of bank examiners’ classifications and the severity of conventional loan default, recent studies have introduced annual debt repayment capacity based on actual financial statements as an alternative indicator of creditworthiness or business performance (Novak and LaDue; Khoju and Barry).

One problem associated with using annual debt repayment capacity as an indicator of creditworthiness is that it only estimates creditworthiness for a single year. Most lenders are interested in creditworthiness over an extended period of time and make credit decisions reflecting this extended period. Loans for real estate, buildings, cattle, and equipment usually are made for terms longer than a single year.

Moreover, the annual debt repayment capacity often vacillates for an individual borrower on an annual basis. This vacillation especially becomes a problem when using debt repayment capacity as an indicator of a borrower’s creditworthiness. For example, a borrower may be identified as creditworthy using a debt repayment capacity ratio, or measure, in one year and less creditworthy the following year— but in actuality the borrower’s innate creditworthiness has not changed. The result stems from annual production, price, and interest rate risk, and the fact that annual debt repayment capacity measures typically do not account for credit reserves or working capital availability. During an adverse economic year, a borrower can utilize credit reserves and/or working capital to smooth cash flow problems caused by production, price, or interest rate risk, and still make all required payments. In essence, a borrower’s creditworthiness status can deteriorate and strengthen over time in response to changes in the basic economic environment or management practices, but should not flip-flop between creditworthiness and less creditworthiness on an annual basis due to normal fluctuations in farm income.

In practice, agricultural lenders generally do not flip-flop credit decisions on an annual basis, but rather deal with normal fluctuations in farm income on an individual borrower and commodity basis. One method used by lenders to objectively ameliorate such annual fluctuations, not only in debt repayment capacity, but more generally in income-based financial ratios and measures, is to calculate and evaluate a borrower’s multiyear average financial ratios and measures. However, statistically based credit scoring models generally have not reflected this lending practice.

Additionally, creditworthiness has been measured historically by loan classification or default which can best (or only) be represented as a qualitative variable (Betubiza and Leatham). However, viewing creditworthiness as a qualitative variable may be less than optimal when debt repayment capacity is used. Typically, debt repayment capacity is derived from a borrower’s financial statements as a continuous, quantitative variable. The benefit of using a continuous, quantitative variable allows the lender to adjust credit levels and financing terms. Furthermore, the continuous, quantitative characteristics of the variable do not preclude it from being converted to a qualitative variable when a yes/no credit decision is warranted.

This study combines two objectives. The first is to introduce smoother, inter-year, extended indicators of creditworthiness and explanatory variables into credit scoring models. The second is to compare a quantitative indicator of creditworthiness to the more conventional qualitative indicator. The first objective is achieved by calculating two- and three-year averages of the debt repayment capacity and explanatory variables, and employing them in credit scoring models. The second objective is achieved by estimating credit scoring models using a measure of debt repayment capacity as the dependent variable and then converting the model’s predicted value of this measure into a binary, qualitative creditworthy classification. These results are compared to those of the more conventional credit scoring model. Conventional credit scoring models utilize a qualitative dependent variable to estimate the model.

The remainder of the article discusses the annual and average indicators of quantitative and qualitative credit-worthiness, defines the explanatory variables, and explains how they are incorporated into credit scoring models using estimation techniques such as ordinary least squares and logistic regression analysis. The data are then discussed, after which the qualitative and quantitative models are empirically estimated and the results are presented and compared. The article concludes with a summary.

Creditworthiness Measures <top>

This study makes a distinction between annual and extended credit-worthiness. Annual debt repayment capacity, measured by the annual coverage ratio,¹ is used to indicate annual creditworthiness. Extended debt repayment capacity, measured by the two-year and three-year averages of the coverage ratio, is used to indicate extended creditworthiness. Extending the creditworthiness indicator, by averaging the coverage ratio, resolves the need to smooth it over time and eliminates some of the inter-year volatility. Furthermore, when two-year and three-year averages of the coverage ratio are employed as the dependent variable in credit scoring models, two-year and three-year averages of the explanatory variables are employed as independent variables, respectively. Smoothing the explanatory variables is a special concern when using income-based financial ratios and measures. The data required by average models should be readily accessible to lenders. Lenders typically maintain historical financial information on existing borrowers and collect two to three years of historical information on new applicants to make a sound credit decision. An exception would be start-up enterprises.

¹The coverage ratio refers to the term debt and capital lease coverage ratio as defined by the Farm Financial Standards Council. Basically, it is calculated as the amount available for debt repayment, without nonfarm income sources, divided by total principal and interest payments. The only deviation from the FFSC’s recommendation is that nonfarm income sources are not included.

In addition to distinguishing between annual and extended creditworthiness, this study also makes a distinction between qualitative and quantitative creditworthiness. In conventional credit evaluation models, the dependent variable usually is represented by a discrete binary variable. In this study, creditworthiness is represented by a borrower’s debt repayment capacity, which is a quantitative variable, yet can be converted to a qualitative variable. The two debt repayment capacity measures, coverage ratio and repayment margin,² recommended by the Farm Financial Standards Council (FFSC), are derived from the borrower’s balance sheet and income statement. The coverage ratio is calculated by dividing the sum of interest and principal scheduled into the funds available for term debt repayment and capital replacement. The repayment margin is calculated by subtracting the sum of interest and principal scheduled from the funds available for term debt repayment and capital replacement.

²The repayment margin refers to the capital replacement and term debt repayment margin as defined by the Farm Financial Standards Council.

The major difference between the two measures is that the coverage ratio is a relative measure, while the repayment margin is an absolute measure, measured in dollars. A shortcoming of the repayment margin is that it is biased toward large farms. Large farms generally have larger repayment margins in absolute dollars than smaller farms, but a larger repayment margin does not necessarily imply greater probability of debt repayment. Since the coverage ratio does not possess this bias and is a relative indicator of a borrower’s probability of debt repayment more suitable for regression analysis, the coverage ratio is selected over the repayment margin.

By itself, the coverage ratio has some unique properties. It focuses directly on the basic characteristic of a creditworthy borrower: the generation of sufficient income to make debt payments. The coverage ratio, as well as creditworthiness, is a function of profitability and debt levels, yet neither explicitly distinguishes between profitability and debt levels. A large (small or negative) coverage ratio should ensure a high (low) probability of debt repayment, and identify a creditworthy (less creditworthy) borrower. However, a small or negative predicted coverage ratio also may indicate a less profitable borrower or a highly leveraged borrower. A large coverage ratio may indicate a more profitable borrower or a low debt borrower.

In addition to distinguishing between creditworthy and less creditworthy borrowers, the coverage ratio—a quantitative indicator of creditworthiness— can assess the degree of creditworthiness. Creditworthy borrowers can be stratified among multiple acceptable credit classifications according to the coverage ratio value for purposes of risk-adjusted pricing, portfolio monitoring, and loan ranking. However, the continuous nature of the coverage ratio does not preclude it from being converted to a binary variable representing credit approval or denial.

A practical issue that needs to be considered, before converting the coverage ratio into a binary variable, is specifying an a priori cut-off value. In previous studies where the coverage ratio is used as an indicator of creditworthiness, a coverage ratio greater (less) than one identifies a borrower as creditworthy (less creditworthy) (Novak and LaDue; Khoju and Barry). An a priori cut-off value greater than one implies all debt obligations and some replacement capital investments were paid from operating income. There is no reason to restrict the cut-off value to one, however. The higher the a priori cut-off value, the more stringent the creditworthiness classification.

This study assigns 1.00, 1.15, and 1.30 as the a priori cut-off levels³ used to distinguish between creditworthy and less creditworthy borrowers. These specific values are selected to demonstrate that creditworthiness models are robust with respect to different a priori cut-off values of the coverage ratio and that no cut-off value is uniquely correct. The cut-off value needs to be determined by each individual modeler or lender according to his/her own needs. Each cut-off value implies a different level of creditworthiness (and capital replacement allowance).

³In this study, the coverage ratio is calculated according to the FFSC recommendations, and cut-off values of the coverage ratio are varied to account for replacement capital investment. An alternative method to account for replacement capital investments is to include an estimate of replacement capital investments in the denominator of the coverage ratio calculations.

Annual and Average Credit Evaluation Models and Estimation Methods <top>

Both the annual and average, and qualitative and quantitative credit evaluation models can be expressed as a linear relationship:

(1)

where

Z refers to the length of the annual or average period;⁴ {i = 1, 2, ..., M } refers to the individual borrowers; {t = 1, 2, ..., T } refers to the period; and refers to the average period. When Z equals 1, t = , and the average model collapses to the annual model. The average dependent variable, and the lagged average dependent variable, are derived by summing the coverage ratio for borrower i over periods Z + t and t, respectively, and dividing by Z. The annual and average current and lagged coverage ratios are used directly in the quantitative models, but are converted to binary variables in the qualitative models. The (N ­ 1) average explanatory variables, are derived by summing the K explanatory variables, X_k,i,t , for borrower i over periods t, and dividing by Z. Again, when Z = 1, an average explanatory variable collapses to an annual explanatory variable. β_k is the corresponding parameter for the annual and average explanatory variable k.

⁴Z equals 1, 2, or 3 in this study.

The credit evaluation model can be expressed as a function of solvency, liquidity, and repayment capacity represented by the debt/asset ratio, current ratio, and coverage ratio, respectively. The emphasis of this study is not to develop a new credit evaluation model, but to compare average and annual creditworthiness measures, as well as quantitative and qualitative credit-worthiness measures, in credit evaluation models. The selected explanatory variables represent a frequently used, parsimonious agricultural credit evaluation model (Miller and LaDue; Miller, Barry, DeVuyst, Lins, and Sherrick).

The parameters in the qualitative credit scoring models are estimated using a logit specification. The logit is based on the logistic regression model specified as:

(2)

where e is the base of natural logarithms and is the probability of the farm being less creditworthy in period given knowledge of from equation (1).

is also known as the logit of the probability of a borrower being creditworthy. The parameters are obtained using the maximum-likelihood estimation method and the likelihood function, because of the inherent nonlinearity of the logistic distribution function (Madalla). To obtain estimates of the unknown parameters, the likelihood function is maximized. The likelihood function expresses the probability of the sample data as a function of the unknown parameters.

The logistic regression estimates the probability of creditworthiness, and borrowers are classified as creditworthy if their estimated probability of credit-worthiness is greater than some prior probability. Following previous investigations, this study uses the total sample population as the prior probability (Miller and LaDue; Turvey and Brown; Novak and LaDue). The prior probabilities are used to classify within- and out-of-sample borrowers. The comparison of out-of-sample predicted borrower classifications with the actual classifications is generally recognized as the most applicable method of validating the models.

The parameters of the quantitative credit scoring models are estimated using ordinary least squares (OLS) regression. The OLS regression models are used to predict the annual or average coverage ratios which, in turn, are used to classify creditworthy borrowers. If a borrower’s predicted coverage ratio is greater than an a priori cut-off level, the borrower is classified as creditworthy. The predicted coverage ratio also can be used as a proxy for the probability of creditworthiness. Some lenders may be more intuitively familiar with a predicted coverage ratio as a probability of creditworthiness than a probability calculated from a logit model.

Data <top>

The data for this study were collected from New York State dairy farms in a program jointly sponsored by Cornell Cooperative Extension and the Department of Agricultural, Resource, and Managerial Economics at the New York State College of Agricultural and Life Sciences, Cornell University. Seventy farms have been Dairy Farm Business Management (DFBM) cooperators from 1985 through 1993, and are analyzed in this study. Such a data set is critical in studying the dynamic effects of creditworthiness. The data include both creditworthy and less creditworthy farms,⁵ and represent a segment of New York State dairy farms which value consistent contribution of financial and management information to DFBM. Additional farm productivity, cost management, and profitability statistics for these farms are summarized in Smith, Knoblauch, and Putnam.

⁵The terminology "less" creditworthy is used instead of "not" creditworthy because it is recognized that the farms have been in operation over a nine-year period and most of them have utilized some form of debt over this period. However, within the sample there are varying degrees of creditworthiness. The sample includes FmHA, Farm Credit System, and various private bank borrowers. Creditworthy to one lender may be less creditworthy to another lender. The data set can be viewed as a compilation of various lenders’ portfolios.

 

Table 1 presents the annual financial ratios for the 70 farms over the 1985­93 sample period. Each financial ratio has been calculated according to FFSC recommendations. Some borrowers reported zero liabilities; therefore, their current ratio and coverage ratio could not be calculated. To retain these borrowers in the sample and avoid values of infinity, the current ratios for these borrowers were given a value of 7, indicating strong liquidity, and the coverage ratios were bounded to a ­4 to +15 interval, depending on the magnitude of the funds available for debt repayment. The bounds of the coverage ratio interval indicate both extremes of debt repayment capacity.

Comparison of Annual and Average Qualitative Classifications <top>

Table 2 summarizes the number of borrowers considered creditworthy over the sample period. The creditworthy classifications are based on annual, two-year average, and three-year average measures of creditworthiness using 1.00, 1.15, and 1.30 cut-off values. As expected, the number of borrowers considered creditworthy decreases as the cut-off value of the coverage ratio increases. Also, the number of borrowers considered creditworthy in each year decreases over time, indicating that some of the borrowers in the sample are becoming less creditworthy. It is probably reasonable to assume that some of the less creditworthy farms will exit the industry if they do not adjust their management strategies and/or if the economic environment surrounding the dairy industry does not improve. Identifying these farms prior to any serious financial problems, such as loan default or loan loss, exemplifies the usefulness of the creditworthiness model.

Though not presented here, in an ancillary analysis, the actual annual and three-year average, and the actual annual and two-year average creditworthy classifications were compared. The results substantiate the earlier hypothesis that actual annual creditworthy classifications based on financial statements fluctuate more over time than actual two- or three-year average creditworthy classifications. For example, seven, four, and ten borrowers in 1991, 1992, and 1993, respectively, are classified as less creditworthy using an annual indicator of creditworthiness, but are classified as creditworthy when using a three-year average indicator of creditworthiness. Conversely, two, one, and three borrowers in 1991, 1992, and 1993, respectively, are classified as creditworthy using an annual indicator of creditworthiness, but are classified as less creditworthy when using a three-year average indicator of creditworthiness. These specific examples are representative of the overall comparison results. In general, the results verify that some farms can fluctuate between creditworthy and less creditworthy on an annual basis, but would maintain a single classification when assessed with an average creditworthiness indicator.

In addition, the two- and three-year average actual classifications were compared. These results, although again not presented here, are slightly different. They indicate that the difference between the two- and three-year classifications is not as significant. In fact, these results are very similar. Most likely, the progressive gains from averaging creditworthiness indicators begin to diminish with each additional period added. However, from this analysis it is not possible to determine which average period is optimal.

Results of Qualitative Creditworthiness Models <top>

Table 3 presents the parameters of the annual, two-year average, and three-year average logistic models with varying a priori cut-off values. All the estimated parameters in each of the models have the expected sign. The prior probabilities, as determined by the sample proportion of classifications, change for each cut-off value, and also are shown in Table 3. As the a priori cut-off values of the coverage ratios increase, the prior probabilities decrease, indicating that fewer farms in the data set met the higher standards of creditworthiness. The two- or three-year average coverage ratios classify a higher proportion of borrowers as creditworthy than the single-year coverage ratio.

Table 4 summarizes the within-sample prediction rates for the annual, two-year average, and three-year average creditworthiness measures. The within-sample prediction rates of the annual, two-year average, and three-year average models, respectively, are: 83.4%, 87.1%, and 90.0% using the 1.00 coverage ratio cut-off value; 84.0%, 86.4%, and 92.9% using the 1.15 coverage ratio cut-off value; and 81.4%, 86.4%, and 85.7% using the 1.30 coverage ratio cut-off value. The within-sample prediction rates of the two-year and three-year average models dominate the annual model’s within-sample prediction rates regardless of selected cut-off value.

Table 5 summarizes the out-of-sample prediction rates for the annual, two-year average, and three-year average creditworthiness measures. The results of the out-of-sample predictions are similar to those of the within-sample predictions. The two-year and three-year models’ prediction rates are better than the annual model’s prediction rates regardless of selected cut-off value. Using a 1.00 coverage ratio cut-off value, the annual model’s prediction rates are 71.4%, 80.0%, and 78.6% for 1991, 1992, and 1993, respectively, while the prediction rates of the two-year and three-year models are 84.3% and 85.7%, respectively. Using a 1.15 coverage ratio cut-off value, the annual model’s prediction rates are 77.1%, 72.9%, and 74.3% for 1991, 1992, and 1993, respectively, while the two-year and three-year models’ prediction rates are 84.3% and 82.9%, respectively. Finally, using a 1.30 coverage ratio cut-off value, the annual model’s prediction rates are 77.1%, 68.6%, and 74.3% for 1991, 1992, and 1993, respectively, while the two-year and three-year models’ prediction rates are 80.0% and 78.6%, respectively.

Any superiority of the two- and three-year average models over each other is not as apparent as the superiority of the average models over the annual models. Sometimes the three-year average model performs better than the two-year average model, while other times the two-year average model performs better than the three-year average model. This statement holds for both within- and out-of-sample predictions. While the prediction rates do not clearly indicate which average period to use, the average prediction results are clearly better than annual prediction results. Some of the improvement in prediction rates can be attributed to the inter-year smoothing of the credit-worthiness indicator and explanatory variables. In addition to improving prediction rates, the average models also extend the period of estimated credit-worthiness.

Note: Numbers in parentheses are P-values.

 

 

Results of Quantitative Creditworthiness Models and Comparison with Results of Qualitative Models <top>

The parameters of the quantitative creditworthiness models, estimated using OLS, are presented in Table 6. All the parameters in the annual and two-year average models exhibit a 5% level of significance, but only the lagged coverage ratio parameter in the three-year average model exhibits a 5% level of significance. The models are evaluated by comparing the out-of-sample actual coverage ratio with the out-of-sample predicted coverage ratio using a correlation coefficient. The annual correlation coefficients are 0.88, 0.57, and 0.50 for 1991, 1992, and 1993, respectively; the two-year and three-year correlation coefficients are 0.82 and 0.77, respectively.

The two-year correlation coefficient of 0.82 for 1991­92 is arguably better than the respective annual correlation coefficients of 0.88 and 0.57 for 1991 and 1992.

^a Correlation between actual and predicted coverage ratio.

Similarly, the three-year correlation coefficient of 0.77 for 1991­93 is arguably better than the annual correlation coefficients of 0.88, 0.57, and 0.50 for 1991, 1992, and 1993, respectively. The two-year and three-year correlation coefficients may not always dominate all the annual correlation coefficients, but they do appear more stable than the annual correlation coefficient for the respective time periods.

It is difficult to draw the same conclusions from a comparison between the two-year and three-year correlation coefficients. The two-year average correlation coefficient is greater than the three-year average correlation coefficient, indicating a closer relationship between out-of-sample predicted and actual coverage ratios, but the three-year average includes an additional year of measured credit-worthiness.

Table 7 summarizes the within-sample predicted coverage ratios from the quantitative model. To make a comparison to the qualitative prediction results, the quantitative predictions are converted to creditworthy and less creditworthy classifications using the comparable a priori cut-off levels. The conversion also can represent a lender’s decision to grant or deny a borrower’s credit request.

Comparison of the results for the within-sample quantitative model versus those for the within-sample qualitative model indicates that neither econometric estimation method dominates the other. The quantitative model does best at predicting the within-sample classifications for: the annual, two-year average, and three-year average models using a 1.00 cut-off value; the two-year average model using a 1.15 cut-off value; and the three-year average model using a 1.30 cut-off value. The qualitative model does best at predicting the within-sample classifications for: the annual and three-year average using a 1.15 cut-off value, and the annual and two-year average using a 1.30 cut-off value (Table 4 versus Table 7). The "true" evaluation of the models, however, is the comparison of out-of-sample prediction rates.

Table 8 summarizes the out-of-sample predicted coverage ratios from the quantitative models. The annual out-of-sample prediction rates indicate that the qualitative models dominate the quantitative models, except when predicting 1991 classifications using a cut-off value of 1.00, and when predicting 1992 classifications using a cut-off value of 1.30 (Table 5 versus Table 8). The two-year and three-year average out-of-sample prediction rates indicate that the qualitative models dominate, or at least equal, the quantitative models for all respective time periods and cut-off values.

Although the qualitative estimation method is generally superior, particularly for the more reliable two- and three-year models, the correct classification rates of the two regression methods are fairly close and the quantitative estimation method provides some promise as an alternative for evaluating individual creditworthiness. It predicts the degree of creditworthiness instead of probability of creditworthiness. This destination may be useful to some lenders for risk-adjusted pricing, establishing financing terms, and combining with nonquantitative borrower attributes for loan decision making. Also, unlike the logistic regression, the OLS regression estimation method can be calculated in most computer spreadsheet packages accessible to most lenders. Possibly, this accessibility might motivate more lenders to use statistically based credit scoring models.

Last, the findings of the quantitative models corroborate the findings of the qualitative models. The average models’ prediction rates dominate the prediction rates of the annual model for each cut-off value, but again, neither average-period model clearly dominates the other.

 

Summary <top>

Credit scoring models using two- and three-year average indicators of creditworthiness achieve higher out-of-sample prediction rates than annual credit scoring models. Thus, the two-year and three-year average creditworthiness indicators and explanatory variables appear more useful than annual credit-worthiness indicators and explanatory variables when employed in credit scoring models. Not only do the average models have higher prediction rates, but they also extend the period of estimated credit-worthiness. Statistically based credit scoring research may need to continue to emulate applied lending practices to improve model performance.

Although the quantitative models (OLS) were not superior to the qualitative models (logistic) in correctly classifying creditworthy and less creditworthy borrowers, the projections of the two- and three-year average coverage ratios using the quantitative estimation procedure provide an alternative method for estimating creditworthiness. Given the ubiquity of computer spreadsheet packages with OLS capabilities, and familiarity of lenders with the coverage ratio, a coverage ratio projection model may be more useful for some lenders.

References <top>

Betubiza, E., and D.J. Leatham. "A Review of Credit Assessment Research and an Annotated Bibliography." Dept. of Agr. Econ., Texas A&M University, College Station, June 1990.

Farm Financial Standards Council. "Financial Guidelines for Agricultural Producers: Recommendations of the Farm Financial Standards Task Force." Revised. American Bankers Association, Washington, DC, July 1995.

Khoju, M.R., and P.J. Barry. "Business Performance-Based Credit Scoring Models: A New Approach to Credit Evaluation." In Regulatory Efficiency and Management Issues Affecting Rural Financial Markets. Federal Reserve Bank of Chicago. Proceedings of North Central Region Project NC-207, Chicago, IL, 4­5 October 1993.

Madalla, G.S. Limited-Dependent and Qualitative Variables in Econometrics. New York: Cambridge University Press, 1983.

Miller, L.H., P. Barry, C. DeVuyst, D.A. Lins, and B.J. Sherrick. "Farmer Mac Credit Risk and Capital Adequacy." Agr. Fin. Rev. 54(1994):66­79.

Miller, L.H., and E.L. LaDue. "Credit Assessment Models for Farm Borrowers: A Logit Analysis." Agr. Fin. Rev. 49(1989):22­36.

Novak, M.P., and E.L. LaDue. "An Analysis of Multiperiod Agricultural Credit Evaluation Models for New York Dairy Farms." Agr. Fin. Rev. 54(1994): 55­65.

Smith, S.F., W.A. Knoblauch, and L.D. Putnam. "Dairy Farm Management Business Summary, New York State, 1993." Res. Bull. No. 94-07, Dept. of Agr., Resour., and Managerial Econ., Cornell University, Ithaca, NY, September 1994.

Turvey, C.G., and R. Brown. "Credit Scoring for Federal Lending Institutions: The Case of Canada’s Farm Credit Corporations." Agr. Fin. Rev. 50(1990):47­57.

 

<top>

 

---

Send questions and comments to Faye Butts fsb1@cornell.edu

This page was last modified on: 02/10/04