This report sheds some light on expected changes in home prices for six cities in two Inland Empire counties. To generate the forecasts, each city was first divided into blocks identified according to the Assessor’s Parcel Numbers. Forecasts of each block’s average prices were then obtained using genetic programming (an artificial intelligence method). Forecasts of home prices belonging to city-blocks with similar prices were then aggregated to produce four city-clusters of hierarchal price-ranges. These forecasts suggest that prices will at best remain sluggish for many geographical locations throughout the end of 2013. Expectations of a housing market rebound remain in the more distant future and should not occur without the unemployment rate decreasing to sustainably lower levels (at least below 6%) first and only if these rates stay there for a while. At the current recovery rate, at least two more years of persistent economic health may pass before home prices start to rebound.
The current economic conditions in the Inland Empire remain at levels that continue to concern residents and policy makers. While economic conditions may be marginally improving throughout most of California, they are not improving in the two Inland Empire Counties of Riverside and San Bernardino as much or as fast. The unemployment rate in the Riverside-San Bernardino-Ontario metro area was 12.8% in June 2012 which remains way above California’s 10.7% unemployment rate. (Although the tri-city unemployment rate was over 14% a year earlier, the rate of reduction seems to be mostly due to increased drop-outs from the labor force.) While marginally improving, current economic conditions remain weak and the slow rebound in home construction activities may result in lower future economic conditions in general – including those of the housing market.
As indicated in last year’s report, improving economic environments transform very slowly into strong housing markets. The 2012 economic environment implies marginally better employment conditions and the housing markets may resume a modest upward trend. Banks may start to extend new real estate loans and new home construction activities may resume.
Similar to last year’s forecasts, this year’s forecasts are produced using genetic programming or GP. (GP is a computerized optimization random search technique that mimics regression modeling.) The computerized algorithm selects the best set of variables out of a large set of explanatory variables to explain prior changes in home prices and produce a forecasting model. GP produces many models that are compared with each other and the ones with the least mean squared regression errors are carefully analyzed to identify the best equations to use in forecasting.
Monthly home-price variations (over the period from January 2010 through June 2012) for each neighborhood are explained by monthly averages of each neighborhood’s home-characteristics (including square footage of homes sold, number of bedrooms, number of bathrooms, ages of those homes, and a dummy variable to account for swimming pools). Exogenously, home prices are explained by six lagged mortgage rates (specifically seven to twelve monthly lags) and seven lagged unemployment rates (specifically six to twelve monthly lags). The mortgage rates (which are identical for all locations) are a proxy variable for cost. The unemployment rates variable (whose data differed from one city to the other but was identical for all neighborhoods within a given city) is a proxy variable for income. Neighborhoods were defined using the Assessor’s Parcel Numbers or APNs. APNs are property identification numbers (much like social security numbers). APNs of adjacent properties typically only differ in a few rightmost digits. To define a “neighborhood”, APN’s of sold properties are truncated to the two or three leftmost digits (depending on the size of the city and number of resulting neighborhoods). The values of the dependent variable to fit and forecast are the average prices of homes sold each month in each neighborhood. When presenting the final forecasts below each city is represented by four clusters identified as C1 for homes sold in neighborhoods at the lowest range of prices, C2 for the second lowest, C3 for the third, and C4 for the cluster with the highest range of prices. Each city is represented by four clusters because the number of APN neighborhoods for each city is different. The forecasts in this report are quarterly home-price changes starting July 2012 through October 2013 for each city’s four clusters.
Before presenting this year’s forecast, a brief evaluation of last year’s forecast seems logical and warranted. The overall mean absolute percentage error (MAPE) in last year’s forecast was 5.7% for the period from July 2011 through June 2012. For the same clusters considered, the lowest 12-month MAPE was 2.5% and the highest was 8.6%. Their graphical representations of forecast accuracies are in Figure 1.
Riverside County Three Cities’ Home Price Forecasts
During the 30 months between January 2010 and June 2012, more than 4,100 homes were sold in Corona. Table 2 has characteristics of those homes through June 2012 within each cluster. C1 represents the lowest quartile of average prices while C4 represents the highest. Figure 2 has the city’s average price history (January 2010 to June 2012) as well as its forecast ranges (of highest, medium, and lowest price tiers) through December 2013. The low tier shown in brown represents expected average prices of the lowest third and the highest tier in green represents the expected average prices of the highest third of all homes sold in the city. (The vertical axis measures prices in $1,000s.) One can easily see that prices are not expected to move much from where they have been over the past two years. Actual (solid blue line), fitted (dotted red line), and expected home prices (highest, medium, and lowest price tiers) in each of the four clusters of Corona through the end of December 2013 are displayed in Figure 3. Figure 4 is a map of Corona’s four clusters’ forecasts at the end of March 2013. Table 3 has a summary of expected quarterly price changes in each cluster through October 2013. The predictions in Table 3 suggest that prices are expected to remain flat in Corona through the end of 2013.
For the same 30 months more than 4,900 homes were sold in Riverside. Characteristics of the homes sold within each cluster between January 2010 and June 2012 are in Table 4. Figure 5 has the city’s average historical and forecasted prices through December 2013. Figure 6 has the independent forecasts of Riverside’s clusters. Figure 7 has the March 2013 forecast. Table 5 has a summary of quarterly price changes in each of the four clusters through October 2013.
For the same period, more than 2,700 homes were sold in Temecula. Characteristics of the homes sold are in Table 6. Figure 8 of the city’s average historical prices and their forecasts suggest little change. Figure 9 of Temecula’s four price clusters suggests relatively more erratic price dynamics however. Figure 10 has the map showing forecasted prices for March 2013. Table 7 contains a summary of expected quarterly changes in prices through October 2013.
San Bernardino County Three Cities’ Home Price Forecasts
Between January 2010 and June 2012, almost 1700 homes were sold in Ontario. Table 8 has the characteristics of the homes sold by cluster. Figure 11 has the city’s average historical prices as well as their forecasts through December 2013. Figure 12 has the independent price forecasts of Ontario’s four clusters. The four clusters March 2013 price forecasts are mapped in Figure 13. Table 9 contains a summary of expected quarterly changes in prices October 2013.
Only about 900 homes were sold in Redlands for the thirty months ending June 2012. Characteristics of the homes sold during the 30 months are in Table 10. Figure 14 has the city’s average historical sale prices starting (January 2010 to June 2012) as well as the forecasts throughout December 2013. Figure 15 has independent forecasts of city’s four price clusters. The March 2013 forecasts are mapped in Figure 16. Table 11 has the forecasted quarterly changes in prices through October 2013.
III. San Bernardino
For the 30 months ending June 2012, more than 3800 homes were sold in San Bernardino. Table 12 has the characteristics of the homes sold there over the past 30 months. The average square footage for homes sold in the city explains the exceptionally low prices there. Figure 17 has the city’s average sale price and its average price forecasts through December 2013. Figure 18 has independent forecasts of the city’s four price clusters. Average forecasted prices for March 2013 are in Figure 19. Table 13 contains a summary of the expected quarterly changes in prices in each of the city’s clusters through October 2013.
Download full report
Maps in this report were prepared by Martin Wong, GIS Analyst, assisted by Serene Ong, GIS Analyst II, The Redlands Institute, University of Redlands.
ABOUT THE INSTITUTE FOR SPATIAL ECONOMIC ANALYSIS (ISEA)
The Institute for Spatial Economic Analysis (ISEA) offers science and research based spatial analysis and forecasts of economic phenomena. ISEA serves regional, national and global communities in their needs to better understand how socio-economic phenomena affect and are affected by their communities and the space they live in. More information and previous ISEA publications can be found at http://isea.redlands.edu/.
You can follow ISEA on twitter @iseaRedlands.