The StreetEasy Price and Rent Indices provide unique insight on price movements within the city’s real estate market. In short, the Indices tell you how much the real estate market has grown (or contracted) over time. In order to see how the market will perform in the future, a forecast is necessary. The StreetEasy Price and Rent Forecasts do precisely that, but it requires several inputs to ensure its accuracy. Most of its inputs provide a clear picture of what is happening now, but median household income is only good for what happened last year. Using the more timely Employment Cost Index, income for the city’s households can be accurately forecasted out one year. With the forecasted income likely to be closer to next year’s income than that value is to the current income value, the income forecast provides a critical input for the StreetEasy Price and Rent Indices forecast.
Median household income (MHI) from the U.S. Census often lags by about three quarters. The Employment Cost Index (ECI) is released at the end of each quarter, and represents the changes in labor costs for businesses in the U.S. As employers’ labor costs are closely related to income for households, the ECI is used as a way to update MHI, and improve the forecast. The Northeast region ECI is used because the region contains New York City.
The forecast of the median household income for the chosen boroughs of New York City is based on a time series model (TSM) for the ECI forecast and that forecast is then used in a least squares regression model to update and forecast the MHI. The TSM is an exponential smoothing state space model, with optimal trend dampening and seasonal adjustments. The models are applied with the ‘forecast’ software package in R (Hyndman, 2014). The regression model is fit with exponentially weighted factors.
The data is from the U.S. Census and the Bureau of Labor Statistics. The data is aggregated into the software package R, where the models are then applied to generate the income forecast. The method was validated by using historical subsets of the data to make forecasts, which were then checked for accuracy against the appropriate historical MHI value. Current forecast accuracy is based on data collected in November 2015.
The accuracy of the forecast was evaluated over several years of existing historical income data for all boroughs. The income forecast error was on average 20 percent smaller than the average yearly change in MHI for each respective borough. With the forecast likely to be closer to the actual MHI in one year than the current MHI is to its value for next year, the forecast appears to be sufficiently accurate to provide a useful MHI for next year.
The average absolute error for the MHI forecast is shown in the table below:
|Area||YoY Change in MHI||Forecast Error|
The median household income for the boroughs of New York City is an important factor in what affects the city’s real estate market. The lag in the MHI, however, reduces its usefulness in understanding where the market may go. The ECI is a way to update the MHI and produce an accurate income forecast. The median household income in Manhattan and Brooklyn has had an average yearly increase or decrease of 4.5 percent and 3.5 percent, respectively, but the forecast has had an average absolute error of 4.1 percent and 2.8 percent, respectively. Forecast errors for the Bronx and Queens are also less than the average absolute change in their respective MHI. With the New York City real estate market reflecting the incredible diversity and dynamism of its inhabitants, the StreetEasy Price and Rent Indices are the best way to understand the market. To have confidence in the forecast of the indices and what they say about where the market may go, means using the most accurate tools possible. The use of the employment cost index in the forecasting of median household income has been shown to be an effective way to build an input for the StreetEasy Price and Rent Indices forecast and that forecast will help people understand what the city’s real estate market may look like next year.
Hyndman, R. with Razbash, S. and Schmidt, D. (2014), “forecast: Forecasting functions for time series and linear models,” R package version 5.6, http://CRAN.R-project.org/package=forecast