Forecast Standard Deviation

• FSD is a statistical measure that represents the probability that the Automated Valuation Model (AVM) value falls within a statistical range of the actual market value, measured against a sales price.
  • Example: If the FSD for an HVE value estimate is 10%, there is a 68% (one standard deviation) probability that the actual market value will fall between +/-10% of the HVE value estimate.
• FSD is expressed as a percentage in decimal form i.e., 13% as .13.
• The lower the FSD, the smaller the error in predicting actual market value.
• FSD is included in your HVE report and can be used to determine your own internal confidence bands.
   

FSD correlates with the HVE Confidence Level

• HVE's Confidence Levels are derived from the FSD and are summarized within high, medium, and low value ranges.
  • High: <.130 FSD (approximately 70% of HVE value estimates)
  • Medium: > .131 and <.200 FSD (approximately 25% of HVE value estimates)
  • Low: > .201 FSD (approximately 5% of HVE value estimates)

FSD is sometimes referred to as "Sigma". However, the generic term "Sigma" can be used when referring to other measures of standard deviation.

• Not all AVMs link Confidence Levels to FSD or other statistical measurements.
• Other AVM's Confidence Levels may be based on less reliable factors such as the number of local properties used in the model's estimate, neighborhood range of values, or other measures that do not correspond precisely to the AVM's performance against the sale price.
• Some AVMs express Confidence Levels in terms of numbers or letters (A, B, C or 1, 2, 3). Inconsistent lettering and numbering across AVMs can lead to confusion and potentially produce results that aren't easily compared or explained. The confusion and inability to connect Confidence Levels to statistically based measurements can cause lenders to guess when making risk management decisions.
• FSD is a statistically based standard deviation which is consistently validated while other standard deviations may be arbitrarily derived based on less reliable factors that are not consistently validated.
• FSDs within HVE are validated to be consistent across all geographic regions through extensive out-of-sample testing with recent Freddie Mac data. For example, a HVE FSD of 10% means the same thing for a property located in North Dakota, South Carolina or California.

• It is important for lenders to understand not only the accuracy of the AVM but also the predictability, or reliability of the Confidence Levels.
• Because FSD is a statistically derived value, it permits full statistical integration of collateral policy with credit policy for lenders when creating their AVM utilization logic.

• The best way to validate Confidence Levels or FSD is to perform an out-of-sample test of property valuations based upon very recent property sale transactions that do not yet appear in public records (for example, purchase-money loans closed within the last 30 days).
• When testing an AVM, a lender should not only validate the accuracy of the model, such as the difference between sales prices and AVM model results, but also validate that the FSD is predictive by measuring whether actual accuracy relative to sale price matches the accuracy predicted by FSD.
• Note: Freddie Mac regularly performs out-of-sample testing of HVE to validate the accuracy and precision of the model estimates as well as the FSD.