Farm-Level and Macroeconomic Determinants of Farm Credit Risk Migration Rates Cesar L. Escalante, Peter J. Barry, Timothy A. Park, and Ebru Demir Abstract Logistic regression techniques for panel data are used to identify factors affecting farm credit transition probabilities. Results indicate that most farm-specific factors do not have adequate explanatory influence on the probability of farm credit risk transition. Class upgrade probabilities are more significantly affected by changes in certain macroeconomic factors, such as economic growth signals (from changes in stock price indexes and farm real estate values) and larger money supply that relax the credit constraint. Increases in interest rates, on the other hand, negatively affect such probabilities. Key words: demographic factors, farm credit risk migration, macroeconomic variables, ordered logit regression, random- effects model, transition probabilities Cesar L. Escalante is assistant professor and Timothy A. Park is associate professor, both in the Department of Agricultural and Applied Economics, University of Georgia. Peter J. Barry is professor and Ebru Demir is former research associate, both in the Department of Agricultural and Consumer Economics, University of Illinois at Urbana-Champaign. The authors wish to thank two anonymous reviewers for their valuable comments and suggestions. Migration analysis, a probability-based measurement concept, has been long employed by such companies as Moodys and Standard and Poor’s (S&P) in evaluating changes in the risk ratings of bonds and other publicly traded securities. The concept has been used more recently to estimate financial stress and/or default rates for commercial, agricultural, and other types of loans (Saunders, 1999; Caouette, Altman, and Narayanan, 1999; Barry, Escalante, and Ellinger, 2002). The migration approach entails tracking an individual borrower’s historic rates of movement among the lender’s credit risk rating classes within a specified time period. These migration rates are used to project the credit quality of loan portfolios according to class upgrades versus downgrades, and derive estimates of probability of loan default or stress rates. Such migration-based measures of credit risk are important inputs in determining lenders’ economic capital requirements under the New Basel Accord (Barry, 2001). Compared to the traditional measurement of historic loan default rates, the migration approach provides richer, broader information on the risk stability and quality of a lender’s loan portfolio, especially when based on more extensive historical data. In agricultural lending, a number of lenders, especially Farm Credit System institutions, have begun to use the migration concept to analyze their loan portfolios, although their data histories generally are less than five years in length and updating of the borrower’s financial
Transcript
Farm-Level and MacroeconomicDeterminants of Farm Credit RiskMigration RatesCesar L. Escalante, Peter J. Barry, Timothy A. Park, and Ebru Demir
Abstract
Logistic regression techniques for paneldata are used to identify factors affectingfarm credit transition probabilities. Results indicate that most farm-specificfactors do not have adequate explanatoryinfluence on the probability of farm creditrisk transition. Class upgradeprobabilities are more significantly affectedby changes in certain macroeconomicfactors, such as economic growth signals(from changes in stock price indexes andfarm real estate values) and larger moneysupply that relax the credit constraint. Increases in interest rates, on the otherhand, negatively affect such probabilities.
Cesar L. Escalante is assistant professor and TimothyA. Park is associate professor, both in the Departmentof Agricultural and Applied Economics, University ofGeorgia. Peter J. Barry is professor and Ebru Demir isformer research associate, both in the Department ofAgricultural and Consumer Economics, University ofIllinois at Urbana-Champaign. The authors wish tothank two anonymous reviewers for their valuablecomments and suggestions.
Migration analysis, a probability-basedmeasurement concept, has been longemployed by such companies as Moodysand Standard and Poor’s (S&P) inevaluating changes in the risk ratings ofbonds and other publicly traded securities. The concept has been used more recentlyto estimate financial stress and/or defaultrates for commercial, agricultural, andother types of loans (Saunders, 1999;Caouette, Altman, and Narayanan, 1999;Barry, Escalante, and Ellinger, 2002).
The migration approach entails trackingan individual borrower’s historic rates ofmovement among the lender’s credit riskrating classes within a specified timeperiod. These migration rates are used toproject the credit quality of loan portfoliosaccording to class upgrades versusdowngrades, and derive estimates ofprobability of loan default or stress rates.
Such migration-based measures of creditrisk are important inputs in determininglenders’ economic capital requirementsunder the New Basel Accord (Barry, 2001). Compared to the traditional measurementof historic loan default rates, the migrationapproach provides richer, broaderinformation on the risk stability andquality of a lender’s loan portfolio,especially when based on more extensivehistorical data.
In agricultural lending, a number oflenders, especially Farm Credit Systeminstitutions, have begun to use themigration concept to analyze their loanportfolios, although their data historiesgenerally are less than five years in lengthand updating of the borrower’s financial
136 Determinants of Farm Credit Risk Migration Rates
data is sporadic. As an alternative datageneration approach, Barry, Escalante,and Ellinger (2002) have utilizedlongitudinal farm-level data to produceestimates of transition probability rates,portfolio upgrades and downgrades, andfinancial stress rates of grain farms inIllinois over a 14-year period. Their studydemonstrates the practical relevance of themigration framework in the assessment ofportfolio quality and its potentialapplication by farm lenders.
This study pursues a more in-depthanalysis of factors that may influence thevolatility of migration rates among farmloans. The analysis focuses on demographicfactors, farm financial/structuralattributes, and macroeconomic conditionsexpected to influence changes in risk classratings over time. The first two factorgroups represent a choice set of variableswhich are mostly within the farmmanager’s control, while the third setrepresents exogenous conditions beyondthe control of individual farmers. Thecredit migration tendencies of sometypes of farms could be more vulnerable tothese cycles than others (Estrella, 2000). This is corroborated by studies ofcorporate bond defaults which haveestablished strong linkages betweendeteriorating economic conditions andtransition to default (Helwege andKleiman, 1997; McDonald and Van deGucht, 1999; Nickell, Perraudin, andVarotto, 1999). Consistent with therecommendations of the Basel Accord,this study also applies the migration andeconometric frameworks to a 10-classcredit rating system.
The following sections review the migrationconcept, discuss the empirical framework,and present the descriptive andeconometric results. A summary andconclusions are provided in the finalsection.
Measuring Migration
Two important considerations in applyingcredit migration analysis are (a) the choice
of classification variable, and (b) the typeof migration measurement. Options forthe classification variable includemeasures of profitability (return on equity),repayment capacity, and the credit score,which is a composite index that usuallyincludes the former measures and otherfinancial factors.
In this study, a farm’s credit score isused to assign farmers to different creditrisk classes. The assignments aredetermined through a credit-scoring modelfor term loans reported by Splett et al.(1994) that is based on financial ratiosrecommended by the Farm FinancialStandards Council.
Table 1 presents the expanded 10-classrating model, recommended under theBasel Accord to more accurately capturedifferences in credit classifications ofborrowers. The class boundaries arebased on the original five-class ratingmodel where, for example, class 1 in thelatter model was divided into classes 1and 2 of the 10-class rating model. Outlier values for the current ratio andthe repayment capacity measures arereplaced by maximum values used byBarry, Escalante, and Ellinger (2002)—i.e., current ratios exceeding the value of 7were assigned the maximum value of 7,while the repayment capacity variablebounds are –1.25 and 0.93.
The farm’s credit score is evaluated usingtwo measurement approaches:
# A year-to-year transition (1 × 1), whichmeasures movements in credit riskratings from one year (n) to the next (n + 1); and
# Three-year average to fourth yeartransition (3 × 1), which measures creditmigration based on a three-year movingaverage of factor data applied to thefourth year. The 3 × 1 approach,informally acknowledged as preferredby farm lenders, allows more gradualmigration than the year-to-yearapproach.
Agricultural Finance Review, Fall 2004 Escalante et al. 137
Table 1. Credit Scoring, Profitability, and Repayment Classification Intervals
LIQUIDITY (Current Ratio) Class 1 > 2.00 Class 2 1.60S2.00 Class 3 1.25S1.60 Class 4 1.00S1.25 Class 5 < 1.00 $$$$ × 0.10 = $$$$SOLVENCY (Equity-Asset Ratio) Class 1 > 0.80 Class 2 0.70S0.80 Class 3 0.60S0.70 Class 4 0.50S0.60 Class 5 < 0.50 $$$$ × 0.35 = $$$$PROFITABILITY (Farm Return on Equity) Class 1 > 0.10 Class 2 0.06S0.10 Class 3 0.04S0.06 Class 4 0.01S0.04 Class 5 < 0.01 $$$$ × 0.10 = $$$$REPAYMENT CAPACITY (Capital Debt-Repayment Margin Ratio) a
Class 1 > 0.75 Class 2 0.50S0.75 Class 3 0.25S0.50 Class 4 0.05S0.25 Class 5 < 0.05 $$$$ × 0.35 = $$$$FINANCIAL EFFICIENCY (Net Farm Income from Operations Ratio) Class 1 > 0.40 Class 2 0.30S0.40 Class 3 0.20S0.30 Class 4 0.10S0.20 Class 5 < 0.10 $$$$ × 0.10 = $$$$
= TOTAL SCORE (Numeric) $$$$$$$$$Credit Score Classes Interval Ranges
Five Credit Classes: Class 1 1.00S1.80Class 2 1.81S2.70Class 3 2.71S3.60Class 4 3.61S4.50Class 5 4.51S5.00
Ten Credit Classes: b Class 1 1.00S1.40Class 2 1.41S1.80Class 3 1.81S2.25Class 4 2.26S2.70Class 5 2.71S3.15Class 6 3.16S3.60Class 7 3.61S4.05Class 8 4.06S4.50Class 9 4.51S4.75Class 10 4.76S5.00
Source: Splett et al. (1994).a New interval ranges for the repayment capacity measure were used in this study since the intervals proposed bySplett et al. (1994) resulted in the heavy concentration of observations in the first class.b The 10 credit classes were derived from the original five credit classes defined by Splett et al. (1994) where class 1 inthe latter classification was split into classes 1 and 2 of the new 10-class approach, and so forth.
138 Determinants of Farm Credit Risk Migration Rates
Proxy Lender and Macro-economic Data Sources
In lieu of scarce lender data, this studyutilizes information from farm financialrecords to estimate credit risks. Thisapproach places greater emphasis onquantitative measures of credit risk anddoes not account for the effects of loancovenants and other risk mitigationstrategies employed by lenders. Incontrast, farm record data could includeborrowers with low credit risk (amongnon-borrowing farms) and high credit risk(accommodated, for example, underfederal financing programs).
The annual farm record data are fromfarms that maintained certified usablefinancial records under the Illinois FarmBusiness Farm Management (FBFM)system during the period 1992 to 2001. The FBFM system has an annualmembership of about 7,000 farmers, butrigorous certification proceduresimplemented by field staff would usuallyresult in much fewer certified farms. Inorder to apply panel data regressiontechniques, the data sets include onlythose farms that consistently maintainedcertified records over the 10-year period. This more stringent requirement produceda total of 116 farms. The FBFM systemprovides demographic and structuralcharacteristics of these farms, as well astheir farm financial performance.
The inclusion of a risk variable calculatedas a three-year moving average and thedetermination of year-to-year migrationrates resulted in eight observations foreach farm under the 1 × 1 migrationapproach. The 3 × 1 method requiredfour annual data points to calculate amigration rate, yielding seven observationsfor each farm.
The macroeconomic measures includedannual averages of Illinois farm real estatevalues from the 2001 annual report of theIllinois Agricultural Statistics Service,long-term interest rates on farm loansfrom the U.S. Department of Agriculture
(USDA), annual changes in the S&P 500,and money supply levels reported by theFederal Reserve Bank of St. Louis.
The Transition ProbabilityMatrices
The average one-period transition matricesfor the 1 × 1 and 3 × 1 measurementapproaches are reported in Table 2. Thevertical axis corresponds to Period 1classes while the horizontal axis showsPeriod 2 classes. Thus, the matrixmeasures the probability that a farmbusiness will migrate from the row classesto the column classes during each period. This probability is calculated as the ratioof the number of farms that migrate to acertain column class (in Period 2) to thetotal number of farms originally classifiedunder a particular row class (in Period 1).
The values along the diagonals in Table 2represent the retention rates, or theprobabilities that farms will remain in thesame class. The off-diagonal elementsrepresent the percentages of upgradesand downgrades in credit classification. Specifically, rightward movements indicatedowngrading while leftward movementsindicate upgrading.
The fixed, finite set of 116 farms evaluatedduring the period 1992S2001 does notaccommodate either new entrants into theclassification system or farms thatterminated operations due to default (i.e.,class 5). Financially distressed farms inclass 5 could either remain in class 5 orexperience an upgrade during the 10-yearperiod.
The migration rates for the five-creditclassification system in Table 2 aregenerally close to values reported by Barry,Escalante, and Ellinger (2002). However,in contrast to the panel data structure ofthis study, their transition probabilitymatrices were constructed using a longertime frame (1985S1998), and themigration rates were separately calculatedusing all available farm observations ineach pair of subsequent time periods.
Agricultural Finance Review, Fall 2004 Escalante et al. 139
Table 2. Average One-Period Transition Matrices for Credit Scores, Five Credit Classes, 1992SSSS2001 (percent)
B. Three-Year Average to 4th Year Transition (3×1):1 74.77 16.51 7.80 0.92 0.002 25.68 42.34 23.87 7.66 0.453 8.60 26.24 41.63 17.19 6.334 3.96 14.85 27.72 27.72 25.745 0.00 4.00 32.00 28.00 36.00
The year-to-year average retention ratesrange from 28% to 73%, while the 3 × 1rates range from 28% to 75%. Consistentwith the results of Barry, Escalante, andEllinger, the retention rates in thisstudy are highest for class 1 borrowers,tend to diminish for the middle-lowercredit risk classes, and slightly increasein class 5.
The retention rates under the 10-creditclassification system (Table 3) aresignificantly lower than those based onfive credit classes. As before, class 1 farmshave greater retention rates compared tofarms in other credit classes. Retentionrates for class 1 farms were calculated at65% and 64% for the 1 × 1 and 3 × 1measurement approaches, respectively. The remainder of the retention rates,however, do not decrease monotonically. In classes 2 to 10, the retention ratesrange from 13% to 32% under the 1 × 1approach, and from 12% to 44% underthe 3 × 1 approach.
Under a seven-bond rating scale (betweenthe 5- and 10-class rating scales usedhere), Moody’s bond rating retention ratesin a one-year transition matrix rangedfrom 56% to 88% over the period1983S1998. A similar matrix developed byS&P for 1981S1996 yielded retention ratesranging from 53% to 89%. In contrast, this
study reports average retention rates ofonly 50% and 32% under the 5- and10-class rating systems, respectively, forthe 1 × 1 measurement approach (Table 4).
In general, studies on bond migrationreflect a greater downgrading thanupgrading of class ratings. For example,Altman and Kao (1992), analyzing firstrating changes among bonds, report that84% of AA bonds downgraded while 17%upgraded. Migration ratings of A bonds,on the other hand, revealed 57%downgrades and 43% upgrades. In thecurrent analysis, this trend only occursunder the 1 × 1 approach, regardless ofcredit classification system used. Specifically, upgrades and downgradesaccount for 47% and 53% (23% and 26%,inclusive of class retention rates),respectively, of the total transition toother credit classes using five creditclasses (Table 4). The percentages forthe 10-class approach are 47% and54% for upgrades and downgrades,respectively.
The trend is reversed for upgrades anddowngrades under the 3 × 1 approach. Perhaps the three-year averaging methodused for determining Period 1 classescushions the impact of volatile andadverse financial conditions on the farm’sinitial rating.
140 Determinants of Farm Credit Risk Migration Rates
Table 3. Average One-Period Transition Matrices for Credit Scores, Ten Credit Classes, 1992SSSS2001 (percent)
Year-to-Year Transition 50.43 31.57 Three-Year Average to 4th Year Transition 48.65 29.31
Upgrades: b
Year-to-Year Transition 23.17 31.79 Three-Year Average to 4th Year Transition 26.23 36.82
Downgrades: c
Year-to-Year Transition 26.40 36.64 Three-Year Average to 4th Year Transition 25.12 33.87
a Summary retention rates were calculated as the average of all diagonal elements of the migration matrices.b Summary upgrade rates were calculated as the average of all transition rates below the diagonal elements ofthe migration matrices.c Summary downgrade rates were calculated as the average of all transition rates above the diagonal elementsof the migration matrices.
Agricultural Finance Review, Fall 2004 Escalante et al. 141
Econometric Framework
The empirical framework utilizes orderedand time-series cross-sectional logitregression techniques performed usingversion 7.0 (special edition) of Statasoftware (Stata Corporation, 2002). Fourversions of the estimating model aredeveloped using the two measurementapproaches (i.e., annual and 3 × 1migrations) for the 5- and 10-creditclassification systems.
Diagnostic test results indicate the needto formulate two separate models for theannual and 3 × 1 migration data sets. TheBreusch-Pagan Lagrangian multiplier(BPLM) test for random effects (with a nullhypothesis that the variance of the unit-specific residual is zero) yieldedcontrasting results for these two data sets. Annual migration data sets for the 5- and10-credit classifications yielded significantBPLM P2 statistics, thus violating thenecessary random-effects assumption. Insignificant Hausman test results furthersuggest the relevance of an ordinaryordered logit regression technique forthese data sets. In contrast, BPLM andHausman test results support therandom-effects model for both the 5- and10-credit classification data sets using the3 × 1 method.
The conceptual form of the estimatingequations is:
migration variable, evaluated on everypair of subsequent periods, that hasvalues of 2 for upgrades, 1 for retentions,and 0 for downgrades; " is the intercept;the vectors Vi t and Wi t (with theircorresponding vectors of regressioncoefficients $$$$1 and $$$$2, respectively)represent structural/demographic andmacroeconomic factors that couldinfluence class migrations; and µi and gi t
are the model’s error terms, with the latterrepresenting the stochastic unit-specificerror components.
Under the random-effects framework, theerror terms are assumed to demonstratethe following properties (Greene, 1993):
E{µiiii} ' 0 and Var{µiiii} ' F2222:::: ,
Cov{giiiitttt, µiiii} ' 0,
Var{giiiitttt% µiiii} ' F2222gggg % F
2222:::: ' F
2222,
Corr{giiiitttt% µiiii, giiiissss % µiiii} ' D.
Logistic regression applies maximum-likelihood estimation after transformingthe dependent variable into a logitvariable, defined as the natural log of theodds that the event of interest will or willnot occur. The ordered logit model is“built around a latent regression in thesame manner as the binomial probitmodel” (Greene, 1993, p. 672). Thus, thecumulative normal probability for a creditupgrade (Yit
* = 2) is specified as a nonlinear(logit) function of demographic andstructural attributes of the farm business(Vi t) and prevailing macroeconomicconditions (Wi t). Moreover, the observedmigration rate denoted by Yit
* in equation(1) depends on a continuous latentvariable (Yit) having various thresholdpoints, as follows:
(2) Y ((((
iiiitttt ' 0 if Yiiiitttt# *1111,
Y ((((
iiiitttt ' 1 if *1111# Yiiiitttt# *2222 ,
Y ((((
iiiitttt ' 2 if Yiiiitttt$ *2222 ,
where *1 and *2 are unknown parametersthat collectively define the range ofvalues for the latent variable (Greene,1993). The *’s are estimated, along withthe unknown $ coefficients of theexplanatory variables.
Assuming that gi t in equation (1) isstandard normally distributed acrossobservations, the probabilities of Yit
* takingvalues of 0, 1, and 2 are:
(3) P (Y ' 0) ' 1
1 % exp jKKKK
kkkk''''1111$kkkkXkkkk & *1111
,
(continued ... )
142 Determinants of Farm Credit Risk Migration Rates
P (Y ' 1) ' 1
1 % exp jKKKK
kkkk''''1111$kkkkXkkkk & *2222
&1
1 % exp jKKKK
kkkk''''1111$kkkkXkkkk & *1111
,
P (Y ' 2) ' 1
1 % exp jKKKK
kkkk''''1111$kkkkXkkkk & *2222
,
where Xk contains regressors Vi t and Wi t,and $k contains their correspondingcoefficients $$$$1 and $$$$2, respectively.
Demographic and Structural/Financial Factors
This analysis considers how farm size,farmland control arrangements, enterprisediversification strategies, and productivityof the existing farm asset complement mayinfluence credit migration. Farm size(SIZE ), measured by gross revenues, couldinfluence the probability of upwardmigration if larger farms experience greaterproduction efficiencies and economies ofscale. These benefits, however, could betempered by higher leverage which createsgreater financial stress.
The contrasting risk-return tradeoffsand liquidity mechanisms, offered byownership through debt financing, shareleasing, and cash leasing, emphasize theimportance of the TENURE variable, whichis defined as the ratio of owned to totaltillable acres of farmland. Ellinger andBarry (1987) have confirmed that highertenure ratios are usually associated withlower accounting rates of return. Shareleasing, on the other hand, offers a highlyrisk-efficient financing option for farmers(Barry et al., 2000). The positivecorrelation between the value of harvestedcrops and the tenant’s rental obligation tothe landowner stabilizes net income,resulting in greater risk-reducing benefitsfor the farm operator. Thus, decisions onfarmland control arrangements couldsignificantly affect the farm’s creditmigration.
Reductions in risk from enterprisediversification could also influence theprobability of upward credit migration. An enterprise diversification index (DIVER )is constructed for each farm using theHerfindahl measure of concentration,calculated as:
H ' jnnnn
iiii''''1111(Shareiiii)
2222.
The index is based on the allocation ofgross farm revenues among the sale ofcrops, livestock, and auxiliary farmservices/products. A fully specializedfarm has an index value of 1, while smallerindex values indicate more diversifiedbusiness portfolios. The influence ofdiversification on the dependent variablewill depend on tradeoffs between riskreduction and high revenue potentialsfrom specialization (Barry, Escalante, andBard, 2001).
The farm’s asset acquisition decisions arereflected by the asset turnover ratio (ATO ),calculated by dividing gross farm revenuesby total farm assets. This measure reflectsthe capability of the farm’s existing assetcomplement to generate revenues. Thegoal is to maximize the assets’ productivecapacity in order to produce optimal levelsof output and sales.
In addition to these structural factors,demographic variables pertaining to thefarm operator’s age (AGE ), geographicallocation (URBINF ), and the soil’sproductivity rating (SOIL ) are also includedin the models. Previous empirical studiescontend that older farmers tend to be morerisk averse (Patrick, Whitaker, and Blake,1980; Lins, Gabriel, and Sonka, 1981),and thus implement more cautiousbusiness plans.
Opportunities for improvements in creditrisk could be greater for farms located nearlarge urban areas. These benefits mightinclude minimization of transaction costsand greater chances at obtaining premiumproduction contracts. In this study, thelocation factor is represented by URBINF,
Agricultural Finance Review, Fall 2004 Escalante et al. 143
an urban influence dummy variable basedon a USDA index where counties areclassified into nine mutually exclusivegroups according to the adjacency to metroareas. This analysis simplifies the indexinto a binary dummy variable equal to 1for counties within metropolitan areas(both large areas with 1 million or moreresidents and small areas with less than 1million residents) as well as for countiesadjacent to large metro areas that eithercontain or do not contain all or part of itsown city of 10,000 or more residents(USDA’s codes 1S4). The variable takes avalue of 0 for non-metropolitan countiesthat are either adjacent to smallermetropolitan areas or are totally rural andisolated communities (USDA’s codes 5S9).
The farm’s soil productivity rating (SOIL ),is an average index representing theinherent productivity of all tillable land ona farm. It reflects the influence on creditmigration of the income-generatingcapacity of crop operations. More stableand higher yield levels are generallyassociated with more productive soil, andthus would positively affect economicperformance.
An income risk (INCRISK ) component,measured as the coefficient of variation(CV) of net farm income, is introduced inthe model. Greater stability of returnsfrom farm revenue sources enablesfarmers to devise effective business plansthat anticipate adjustments in the farm’sliquidity and profitability conditions. Ultimately, better farm financialperformance results in greater likelihood ofimprovements in credit risk ratings.
Macroeconomic Variables
The success or failure of a farm businessdoes not solely depend on the farm’sability to implement growth-enhancingand risk-reducing business plans. Macroeconomic forces, beyond the farmer’scontrol, could significantly influence theeffectiveness of such business strategies. This analysis considers severalmacroeconomic measures related toeconomic growth, lending conditions,
investor expectations, and price levelchanges that are expected to influencethe credit risk migration trends of farmbusinesses.
Among alternative proxy measures foreconomic growth activity, the annualgrowth rates of farm real estate values(FLGRWTH ) provide a comprehensiveindication of growth both within the farmindustry and the economy in general. Variation in the growth of farm real estateprices does not only depend on farm-related conditions such as changinggovernment farm policies, productionrisks, and farm credit conditions, but alsoon non-farm investment opportunitiesdictated by the economy’s demands forcommercial, residential, and recreationalfacilities, among others.
The availability and cost of credit are alsoimportant determinants of the likelihoodof upward migration. The annual growthrate of the economy’s monetary stock(MNYGRWTH ) is used in this analysis toreflect changes in credit availabilityconditions. Bankruptcy studies haveobserved that the majority of businessfailures among small firms occur duringtight money conditions when lendersusually resort to small business “credit-rationing” to protect their loan portfolios(Altman, 2001).
Changes in credit costs are represented bythe annual change in interest rates foragricultural mortgage (long-term) loans(AGRATES ). Interest rate adjustment isnormally among the policy options used bythe Federal Open Market Committee toachieve certain economic goals. Forinstance, the Federal Reserve Board’saggressive rate-cutting campaign fromJanuary 2001 to June 2003 was designedto stimulate greater economic activity fromthe business, consumer, and marketsectors of the economy. Compared toshort-term interest rates that are easilyaffected by changes in the federal fundsrate, longer-term borrowing rates follow amore complicated adjustment processinvolving other indicators, such asspeculative and precautionary factors.
144 Determinants of Farm Credit Risk Migration Rates
Table 5. Results of Ordered and Random Effects Logit Regression, Multinomial Dependent Variable
Note: Single, double, and triple asterisks (*) denote significance at the 90%, 95%, and 99% levels, respectively.
Finally, credit risk migration could also beaffected by the general economic outlookas reflected in both the prices being paidfor holding financial assets, such asstocks, and the risk premium investors arewilling to pay for keeping riskier financialassets (Altman, 2001). The S&P 500 indexof stock prices is used as a proxy for theoverall stock market performance. Annualchanges in the stock price index (SPCHG )reflect changes in the investors’ demandfor holding stocks.
Econometric Results
Except for the income risk variable, thedependent variable is regressed againstthe two-year and four-year averages of theannual values of the structural anddemographic variables under the annualand 3×1 migration frameworks,respectively. One-year lagged growth ratemeasures for the macroeconomic variablesare used for the annual migration datasets. In the 3×1 migration data sets, theequivalent growth rate measures the
average growth rate for every four-yearperiod.
The models’ coefficients provideunambiguous indications of changes in theprobability of moving from the lowest tothe next highest categories, and vice versa(upgrades and downgrades), in addition toimportant information on the model’sexplanatory power and the statisticalsignificance of each individual independentvariable. The regressors’ directional effectscan be discerned, however, from estimatesof their marginal effects. The followingsections discuss the significance of certainvariables and their directional effects ineach category of the dependent variable.
Significant Determinants
Table 5 reports the coefficient estimatesand the resulting Z-statistics for thesignificance tests for the four versions ofthe model. A positive (negative) coefficientfor a regressor suggests it increases(decreases) the odds of a credit classupgrade.
Agricultural Finance Review, Fall 2004 Escalante et al. 145
Table 5. ExtendedThree-Year Average to 4th Year Transition
Among the two groups of regressors, none ofthe eight demographic, financial/structuralvariables had a significant influence on theprobability of credit migration in the 3×1random effects model using five creditclasses. This result could reflect thedistributional characteristics of the dataset—i.e., homogeneous demographic andstructural attributes may not yield enoughvariability to significantly affect creditmigration probabilities. Moreover, certainvariables could have dual, offsetting effectson the dependent variable. For example,the greater production capacity of largerfarms could favor upgrades, whileshortfalls in production efficiency couldlead to downgrades.
The other three models (ordered logitmodels for 5-class and 10-class annualmigration and 3×1 random effects modelusing 10 credit classes) produced at mosttwo significant demographic, structural/financial variables. The diversification(DIVER ) variable’s significant negativecoefficient in the 3×1 method, 10 classesmodel suggests that increasingspecialization of farm enterprises could
lead to greater probability of classdowngrades. This result aptly describesthe regional distribution of farm operationsin Illinois where the relatively lessproductive soil profiles of the southerncounties create a greater necessity todiversify farm enterprises. In contrast, thehighly productive soils in the north andcentral regions normally allow their farmsto specialize in corn, soybean, and wheatproduction. However, this study’s sampleperiod captures episodes of steadilydeclining grain prices as a result of supplyoverstock in the mid-1990s while federalprograms wavered from providing risk-reducing countercyclical subsidies to fixed,decoupled payments. Hence, the morediversified crop-livestock farms in lessproductive regions have been moreresilient and more likely to realize upwardmobility in credit risk ratings.
Asset turnover (ATO ) is significant andpositive in both the 5-class and 10-classdata sets, suggesting that farms betterable to increase the productive capacity oftheir farm asset complements are morelikely to experience rating upgrades.
146 Determinants of Farm Credit Risk Migration Rates
Table 6. Marginal Effects of Significant Explanatory Variables5 Credit Classes
B. Three-Year Average to 4th Year Transition (3× 1): DIVER (diversification index) — — — FLGRWTH (farmland value growth, %) !26.94238 !1.23399 28.17636 MNYGRWTH (money supply growth, %) !2.64853 !0.12130 2.76983 SPCHG (S&P 500 change, %) !0.77385 !0.03544 0.80929 AGRATES (change in ag LT interest rates, %) 4.40746 0.20187 !4.60932
AGE also has the same effect in the10-class data set using the annualmigration method. Its positive coefficientsuggests older, more experienced farmersare more likely to experience ratingupgrades. Moreover, older farmers aremore likely to maintain favorable,affordable debt loads that have beengradually retired over the years.
The overall weak, insignificant impact ofthe farm’s structural and demographicprofile could imply that such attributes areemphasized more for making loandecisions and defining loan covenants. Once the loan is granted and serviced,these factors become less relevant indetermining credit migrations.
A major result of this analysis is the stronginfluence of macroeconomic variables onthe dependent variable. Changes in thevalues of money supply (MNYGRWTH ),farm real estate (FLGRWTH ), and stockindex (SPCHG ) are consistently significantamong the macroeconomic variables in allfour estimating equations. High growthrates in money supply (MNYGRWTH ) relaxthe credit availability constraint and allowfarmers to undertake strategies thatincrease the likelihood of credit upgrades. FLGRWTH has a similar positive effect onthe dependent variable. Increases in farmreal estate values point to a flourishingfarm economy, thereby increasing the
probability of credit upgrades. Thepositive sign of SPCHG is consistent withthe expectation that a growing stockmarket index could influence theprobability of upgrades.
AGRATES, a measure of the changes inagricultural mortgage rates, is significantand negatively signed in both modelversions using the 3×1 measurementapproach. Higher interest rates canincrease financial risks for indebtedfarmers and lead to downgrades in riskratings.
Directional Effects
The directional effects are more explicitlygiven by the marginal effects of thesignificant variables in Table 6 (StataCorporation, 2002). Among thedemographic and structural variables, theprobability of experiencing a downgrade ismore sensitive to unit changes in assetturnover (ATO ) than to similar incrementsin operator’s age (AGE ). Specifically, thelikelihood of a downgrade decreases by arange of 0.13446 to 0.16002 due to a unitincrease in ATO, while the equivalentchange for AGE is 0.00272. Theprobabilities of retentions and upgradesincrease for every unit change in each ofthese two financial variables, with ATOyielding the larger effects.
Agricultural Finance Review, Fall 2004 Escalante et al. 147
B. Three-Year Average to 4th Year Transition (3× 1): DIVER (diversification index) 0.25208 0.01141 !0.26348 FLGRWTH (farmland value growth, %) !31.52756 !1.42648 32.95404 MNYGRWTH (money supply growth, %) !3.15944 !0.14295 3.30239 SPCHG (S&P 500 change, %) !0.92407 !0.04181 0.96588 AGRATES (change in ag LT interest rates, %) 4.19605 0.18985 !4.38590
The positively signed macroeconomicvariables (SPCHG, MNYGRWTH, andFLGRWTH ) in Table 5 consistently havemarginal effects revealing a negative and apositive effect on the probability of a classdowngrade and upgrade, respectively, inTable 6. Their directional effects on theretention probability, however, are nothomogeneous. These variables hadnegative effects on the retentionprobability in the two models under the3×1 method. The equivalent effect in theannual migration models is positive.
AGRATES, a negatively signed regressorin Table 5, has positive marginal effectson downgrade and retention probabilities,while its marginal effect on the probabilityof upgrades is negative. Among all themacroeconomic variables, farm real estategrowth yielded the strongest marginaleffects. This variable consistentlynegatively influences downgradeprobabilities. Notably, in the 3×1models, this effect ranges from 26.9 to31.5 times. In the same models, thisvariable’s effect on upgrade probabilitiesis positive and ranges from 28.2 to 33.0times. The results for retentionprobabilities are mixed, with positiveprobability effects in the annual modelsand negative effects in the 3×1 models. The variable’s effect, however, isconsistently positive for the probability ofa class upgrade.
Summary and Conclusions
This study introduces two newperspectives in understanding theapplication of the migration model to farmcredit risk analysis, i.e., the expansion ofthe credit classification system from 5 to10 classes and possible determinants ofcredit migration probabilities. Consistentwith the recommendation of the BaselAccord, an expanded 10-class version ofthe 5-class credit rating system isintroduced to determine its impact ontransition probabilities. The econometricanalysis also considers farm-level as wellas macro factors that are both within andbeyond the farm manager’s control.
The migration matrices obtained in thisstudy reflect trends of lower class retentionrates and highly volatile transitionprobabilities compared to results obtainedfor bonds and other publicly tradedsecurities (Barry, Escalante, and Ellinger,2002; Altman and Kao, 1992), althoughthe lenders’ subjective inputs in the loandecision process have not been consideredhere. Nonetheless, this result is consistentwith the riskier nature of farmingoperations that are easily more susceptibleto seasonal fluctuations in weather andmarket conditions than firms belonging toother industries. Notably, the shift fromthe conventional 5-credit classificationsystem to an expanded 10-class approach
148 Determinants of Farm Credit Risk Migration Rates
produced a greater incidence of classmigrations with higher overall rates ofupgrades and downgrades than retentionrates.
The econometric results under the 5- and10-credit rating scales were, however,more consistent with each other. Ingeneral, this analysis demonstrates thatmost farm-specific factors do not haveadequate explanatory influence on theprobability of credit risk transition. Thehomogeneity of farm conditions or theoffsetting interaction effects of certainfactors could have minimized theimportance of the farms’ demographicand structural attributes.
The more compelling result is thedominance of macroeconomic factors onthe probability of credit migration. Increases in stock price indexes and farmreal estate values both signal a growingeconomy through aggressive investmentactivities and expansive projectdevelopments. They are thus associatedwith the greater likelihood of classupgrades. The relaxation of the creditconstraint through increments in themoney supply level strengthens thelikelihood of upgrades, while increases ininterest rates make downgrades morelikely.
Future research applying the migrationframework to farm finance could expandthe analytical model used here to accountfor other factors directly or indirectlyaffecting transition probability rates suchas weather, irrigation systems,technological change, and social capital.
References
Altman, E. I. Bankruptcy, Credit Risk, andHigh Yield Junk Bonds. Cambridge, MA:Basil Blackwell, Inc., 2001.
Altman, E. I., and D. L. Kao. “TheImplications of Corporate Bond RatingsDraft.” Finan. Analysts J. 48,3(May-June 1992):64S75.
Barry, P. J. “Modern Capital Managementby Financial Institutions: Implicationsfor Agricultural Lenders.” Agr. Fin. Rev.61,2(Fall 2001):103S122.
Barry, P. J., C. L. Escalante, and S. K. Bard.“Economic Risk and the StructuralCharacteristics of Farm Businesses.” Agr. Fin. Rev. 61,1(Spring 2001):73S86.
Barry, P. J., C. L. Escalante, and P. N.Ellinger. “Credit Risk Migration Analysisof Farm Businesses.” Agr. Fin. Rev.62,1(Spring 2002):1S11.
Barry, P. J., L. M. Moss, N. L. Sotomayor,and C. L. Escalante. “Lease Pricing forFarm Real Estate.” Rev. Agr. Econ.22(Spring/Summer 2000):2S16.
Caouette, J. B., E. I. Altman, and P.Narayanan. Managing Credit Risk. New York: John Wiley & Sons, 1999.
Ellinger, P. N., and P. J. Barry. “TheEffects of Tenure Position on FarmProfitability and Solvency: AnApplication to Illinois Farms.” Agr. Fin.Rev. 47(1987):106S118.
Estrella, A. “Credit Ratings andComplementary Sources of CreditQuality Information.” Basel Committeeon Banking Supervision Working Papers3(August 2000):132S150.
Federal Reserve Bank of St. Louis. “Economic Data—FRED II.” Online. Available at http://research.stlouisfed.org/fred2/. [Accessed August 6, 2003.]
Helwege, J., and P. Kleiman. “Under-standing Aggregate Default Rates ofHigh-Yield Bonds.” J. Fixed Income7,1(June 1997):55S61.
Lins, D. A., S. C. Gabriel, and S. T. Sonka. “An Analysis of the Risk Aversion ofFarm Operators: An Asset PortfolioApproach.” West. J. Agr. Econ. 6,1(July1981):15S29.
Agricultural Finance Review, Fall 2004 Escalante et al. 149
McDonald, C. G., and L. M. Van de Gucht. “High-Yield Bond Default and CallRisks.” Rev. Econ. and Statis.81,3(1999):409S419.
Nickell, P., W. Perraudin, and S. Varotto. Rating versus Equity-Based Credit RiskModelling: An Empirical Analysis. London: Bank of England, BirbeckCollege, 1999.
Patrick, G. F., S. H. Whitaker, and B. F.Blake. “Farmers’ Goals and RiskAversion: Some Preliminary Analyses.” In Risk Analysis in Agriculture: Researchand Educational Developments. Pub. No.AE-4492, Dept. of Agr. Econ., Universityof Illinois, June 1980.
Saunders, A. Credit Risk Measurement. New York: John Wiley & Sons, 1999.
Splett, N. S., P. J. Barry, B. L. Dixon, andP. N. Ellinger. “A Joint Experience andStatistical Approach to Credit Scoring.” Agr. Fin. Rev. 54(1994):39S54.
Standard and Poor’s. “S&P Indices.” Online. Available at http://www2.standardandpoors.com/NASApp/cs/ContentServer?pagename=sp/Page/IndicesIndexPg&r=1&b=4&s=6&ig=51&l=EN&i=56&xcd=500. [Accessed August14, 2003.]
Stata Corporation. Stat/SE 7.0 forWindows 98/95/NT. College Station,TX, 2002.
U.S. Department of Agriculture, EconomicResearch Service. “Agricultural Incomeand Finance.” USDA/ERS, Washington,DC, February 2002. Online. Available athttp://jan.mannlib.cornell.edu/reports/erssor/economics/ais-bb/2002/. [Accessed August 14, 2003.]
U.S. Department of Agriculture, IllinoisAgricultural Statistics Service. “2001Illinois Annual Summary.” Online. Available at http://www.agstats.state.il.us/annual/2001/01095.htm. [Accessed August 14, 2003.]
