Fannie Mae

Other

Mark Y An

Andre H Gao

Eric Rosenblatt

Hong Gao

Cristian A De Ritis

Linghua Peng

Market based data cleaning for mitigation of idiosyncratic errors in transaction data used for property valuation. The market based data cleaning technique helps to ensure that the most accurate record is retained for a transaction when duplicate records are found, by ensuring that the retained record is the most consistent with other transactions of the same property, the local market trend, and neighborhood market. Algorithms accommodate the adoption of a value as a representative single record for a transaction where multiple records are present for a transaction. Following duplicate removal, market based data cleaning further eliminates erroneous records by eliminating transaction outliers, also preferably based upon the local market trend, along with all of the remaining transactions for each given property.

Market based data cleaning

A method and system for evaluating the performance of a study group of loans relative to the performance of a control group of loans is presented. The described method includes determining a study group and a control group from a portfolio of individual loans, determining a set of control variables associated with each of the individual loans included in the study group and the control group, partitioning the control group into one or more cells based on the set of control variables, calculating an estimated performance value for each of the individual loans in the study group based on the cells; and calculating a performance difference for the study group. The performance difference has a standard normal distribution with zero mean which allows statistical significance to be estimated using standard probability tables. The control group may either include or exclude the individual loans in the study group to eliminate inaccuracies.

Method and system for assessing loan credit risk and performance

A method and machine-readable media are provided for estimating property values. An exemplary method includes obtaining a first estimate of a property from a first property valuation database, the first estimate exhibiting price tier effect bias. The method further includes deriving a price tier effect adjustment factor based on a price tier associated with the property and purchase transaction data. The method further includes adjusting the first estimate to reduce the price tier effect bias using price tier effect adjustment factor, thereby providing a second estimate having reduced price tier effect bias.

Method and system for automated property valuation adjustment

Property valuation that uses multiple transactions in predicting a value for a property. Instead of choosing only one prior transaction, valuation uses multiple transactions for a given property to provide a more accurate mark-to-market value. Preferably, a weighted combination of mark-to-market values provided by an HPI and individual ones of the multiple transaction records provide a predicted value for the given property. Weighting factors recency and transaction type to correct for both sources of potential inaccuracy.

Multiple transaction property valuation

A method of determining a probability of an adverse event in connection with a loan (such as, delinquency, default, or prepayment) based on a predetermined set of loan parameters comprises determining whether loan variable data is available for each of the loan parameters. If loan variable data is available for each of the loan parameters, the probability of delinquency within a period of time is estimated as a function of the loan variable data. If loan variable data is not available, the probability of delinquency within a period of time is estimated by imputing the loan variable data that is not available.

A method and system are provided removing systematic bias from property value predictions obtained using automated property valuation models. The systematic bias is removed by deriving monthly adjustment factors that are location-specific. The monthly adjustment factors are obtained by comparing newly obtained purchase transaction data to model-generated valuations of properties in a base valuation database. The newly obtained purchase transactions may be obtained from newly obtained data on recently completed purchase transactions and loan application data on future purchase transactions. The monthly adjustment factors are applied to the property valuations in the base valuation database to obtain adjusted property values having reduced bias and thus improved accuracy.

A repeated transaction index (RTI) for estimation of a home price index (HPI) that controls for systematic bias. Purchase transaction records are used to estimate an HPI for areas large enough to have sufficient transaction data. Bias is removed from non-purchase transaction records using the HPI. All available data is used to estimate a localized HPI that pertains to smaller geographical areas (e.g., zip codes, neighborhoods, etc.) after correction of the biases in the non-purchase transaction records.

Trunk branch repeated transaction index for property valuation

Confidence scoring entails receipt of a predicted property value for a given property, where the predicted property value is determined based upon application of a property value prediction model. A set of values corresponding to input factors is determined for the property value prediction model, and model parameters are estimated based upon those values. A confidence score for the predicted property value is then calculated using a confidence scoring model and the model parameters. The confidence score is preferably provided as an accuracy probability. This may further be in the form of a probability that the predicted property value has an error within a specified range, such as a specified range expressed in terms of a percentage error.

Growing community of inventors

North Potomac, MD, United States of America

Mark Y An