At ResiShares, we spend all of our time investigating and operating single family residential real estate. We come from quantitative, data-intensive backgrounds and we build systematic models to power our business. In our daily research, we are surrounded by data, models and analysis about all aspects of houses, neighborhoods, and markets. Systematic research like this aims to impose quantitative rigor on the formulation of investment theses, in order to avoid the common human pitfall of cherry picking data to reinforce our favored narratives.
At the same time, our brains have evolved to tell stories; not to precisely calculate complex interactions of mathematical variables. One important way that we humans distill data into digestible narratives is to compare what we are trying to describe in the present with something similar that we observed in the past, and whose outcome is known. In this way, Austin in 2016 becomes “the next San Francisco.” Salt Lake City in 2019 becomes “the next Denver.” The Trended Real Estate Comparison Score (T-RECS) is our effort to systematically relate places through time by following the data wherever it leads. This approach captures the richness and potential of a place through the human lens of comparison, while imposing the discipline of systematic analysis to avoid errors in judgment.
Our data process continuously ingests, cleans, and normalizes data from dozens of sources including county public records, government agencies, data vendors, and internal operating data. Our research process transforms, combines, and isolates the value of these data points and systematically updates our machine learning models on a monthly cadence. Our modeling efforts are cross sectional examinations across markets, neighborhoods, and individual assets. In all cases we target 3-year relative price growth. The 3-year target is short enough to retain predictive signal strength from our model, but long enough to materially impact returns on a typical-duration real estate investment.
We chose to work with relative, rather than absolute price growth to isolate the fundamental drivers of secular outperformance across real estate markets from the overall macroeconomic backdrop. Traditional forecast models offer point-estimates of future values with what we believe to be false precision. In a trending market, these models often land admirably close to observed reality, but they can also chase momentum and miss major inflection points, rendering them difficult to use practically to inform purchase and investment decisions. A relative approach explicitly avoids predicting a specific level of future price appreciation, but instead focuses on evaluating the potential for a given market to outperform other markets.
T-RECS (Trended Real Estate Comparison Score) is a statistical measure of comparison between the most populous housing markets (Core Based Statistical Areas, or CBSAs) in the present day, and other CBSAs at historical points in time. We compare cross sectionally normalized characteristics that our market forecasting model finds predictive of future appreciation as well as descriptive factors that underpin the unique characteristics of each CBSA. A selection of the more human-interpretable characteristics are shown in a table, while others are combined into 3 scores for readability. We consider the top 200 CBSAs, over multiple decades. The data is sourced from public records, governmental surveys, a number of data vendors, and internal/proprietary data and models.
The comparable CBSAs are listed in the lower right table, including the date at which they are most similar. To generate this list, we treat each CBSA-Month as a unique data point with N normalized characteristics. We then consider a distance metric on this N-dimensional space which weighs the CBSA specific characteristics, and also a lesser weight on the overall economic climate of the compared months. We find the closest 5 neighbors by this distance metric, removing duplicate months of the same CBSA. These comparables serve as a guide of some of the range of outcomes that the selected CBSA may encounter.
The historical and forecast percentile rank of the chosen CBSA’s housing market relative to other CBSAs is shown on the upper right table. Each point represents the relative appreciation for that particular year. For example a line going from the 40th percentile in 2022 to the 60th percentile in 2023 and then to the 80th percentile in 2024, represents a CBSA out-appreciating 40% of the nation in 2022, 60% in 2023, and 80% in 2024. The cumulative expected percent rank of appreciation over the next 3 years can be seen above the chart.