The Weiss Analytics (WA) forecast integrates the unique Weiss house level indexes with local, regional, and national real estate indicators using state of the art machine learning algorithms. This method allows WA to create highly predictive house level price change forecasts.
WA regularly performs large scale out-of-sample tests on its forecasts. The most recent test, completed in March 2019, included approximately 500,000 properties covering the period from 2004 through 2018 in 13 metropolitan areas.
The Weiss one-year forecast predicts how much a property’s value will rise or fall over the following year. Its accuracy can be measured by comparing it to the change in the Weiss index computed a year later.
The median signed error in the most recent test was 0.1%, indicating virtually no bias in the forecasts.
Another way to measure forecast accuracy is how well it predicts the direction of property value changes. Since home prices have inertia, it is relatively easy to predict price changes when appreciation or depreciation continues year-over-year. However, a forecast is more important to mortgage investors in times of high market volatility. For example, it is more important to know that home values are beginning to decrease in a certain neighborhood, than whether the magnitude of an ongoing decline is 5% or 9%.
WA defines three direction ranges of price trends, up, when prices rise more than 3% in a year, down, when home prices change are less than 1%, and flat, when the change is between 1% and 3%. These are chosen to be within 1% of the 2% average yearly inflation rate that has existed for several decades.
WA compares the forecast direction, up, down or flat, to the following year’s actual direction. The Weiss forecast is highly accurate in predicting changes in market direction.
WA also produces a longer-term, three-year forecast. Here are results for the predictive accuracy of this forecast versus actual 3-year home price appreciation. We choose the “flat” range to span the range 3% to 9%, three times the one-year range
Overall accuracy for the 3-year forecast when values are declining is very good, with predictions in the incorrect “up” range only 8% of the time. In the yearly breakdown, the forecast is overwhelmingly correct through 2009. In 2010 many predictions are in the “flat” range. Beyond that, there are few down predictions, but the absolute number of homes with declining values is also very low.
Overall accuracy for the 3-year forecast for rising values is lower than for declining values, but still very good. The yearly breakdown is overwhelmingly correct in all years except 2004, where the final phase of the housing boom was not fully predicted
Akron OH, Atlanta GA, Boston MA, Cedar Rapids IA, Denver CO, Miami FL, Oklahoma City OK, Omaha NE, Philadelphia PA, Pittsburgh PA, Portland OR, Rochester NY, and San Francisco CA