On June 8 2020 the National Bureau of Economic Research (NBER) announced a recession that began in February 2020 and ended two months later. It was the shortest recession on records dating to 1854, according to NBER's list of U.S. business cycle expansions and contractions.
Although the COVID-19 recession proved short-lived, and while recessions may not necessarily get followed by real estate price downturns, it prompted a study I did at the time, on what economic variables served the best statistical prediction of real estate price downturns during the Global Financial Crisis. The latter study has been extremely valuable in my private investment decision making. I lay out some of the findings below.
Real Estate Downturns Post the Global Financial Crisis
Below are the peak-to-bottom real estate price drops for U.S. states post the Global Financial Crisis, based on Federal Reserve Bank of St. Louis data. It took a median 4 years and 1 quarter to reach price bottom, with median peak date in Q4 2007 and median bottom date in Q1 2012.
The largest magnitude price drops observed were in the states of Nevada, Arizona, Florida, California in the range of 41-56%. The smallest magnitude price drops observed were in states including North Dakota, Iowa, Texas, Kentucky among others at mere 0-5%. My goal was to understand what prompted such significant discrepancy in downside performance.
Market Valuation Metrics as Predictors of Downside Risk
The Case for Valuation Metrics, From Finance
The concept of valuation metrics is customary in finance, as ratios or models that give an investor an idea of what a particular company may be worth. Such may be aggregated at the market level as well, providing a measure of 'market valuation', how a market measures as an aggregate relative to intrinsic value. Perhaps the most well known is the Price/Earnings ratio (P/E), though further metrics and approaches are utilized as well, such as Tobin's Q, Shiller CAPE, Fed Model, Price/Book value among others.
A study by John P. Hussman, Ph.D, from his article called 'Measuring the Bubble' reveals correlation of various valuation metrics to actual subsequent stock market total returns. It showcases the capacity to devise custom market valuation metrics, which may perform to a higher correlation to subsequent price changes, such as Hussman's own Nonfinancial market cap/Corporate gross value-added and Margin-Adjusted CAPE metrics.
Most Overvalued Countries
A cross-country study on Housing Bubble Risks published by Niraj Shah of Bloomberg Economics, takes into consideration four factors: House Price-Rent Ratio, House Price-Income Ratio, Real House Prices, Credit of Households (% of GDP).
It showed Canada, Scandinavia, Oceania, UK significantly overvalued by both House Price-Rent Ratio and House Price-Income Ratio.
Ingo Winzer Pre-2007 Housing Bubble Warning
Ingo Winzer of Local Market Monitor publishes metrics on various U.S. real estate markets.
In 2005 he was quoted by CNNMoney as below, with a detailed table published the following year:
"I think Americans are not well aware that many markets are risky," says Ingo Winzer, president of Local Market Monitor … Winzer considers real estate "very risky right now.“ …"Bubbles do tend to last longer than most people expect," he says, "and end quicker."
That real estate was overvalued at the time, was in fact measurable.
Real Estate Market Valuation by U.S. State at Peak Pre-Global Financial Crisis
At the beginning of COVID, I looked at the following variables as possible predictors or informants of price downside post market peak:
foreclosures activity data: source ATTOM data solutions
public real estate (REITs) price performance: iShares Residential Real Estate Capped ETF
price volatility by region (relative standard deviation), risk-adjusted returns by region (Sharpe ratio)
correlation with overall market by region, autocorrelation by region
population by region (natural increase, domestic migration, international migration), income by region
valuations based on historical price and income time series.
I performed a back study statistically predicting price downturns post the Global Financial Crisis. Valuations based on historical price and income time series, proved extremely predictive.
Valuation = % Deviation from historical Price/Income Ratio
Specifically, percentage deviation from historical price/income ratios in each region, when taken against a well-selected moving average window, showed 83-87% Pearson correlation vs subsequent actual price drops, until bottom.
Regression was performed at state level, exact correlation magnitude dependent on price data set and whether the District of Columbia included or not.
Below is a map of real estate market valuations by U.S. state. Correlation measured is between valuation percentage as seen on the map and the exact price downturns peak-to-bottom post the Global Financial Crisis. One can see that the 41-56% downturn states of Nevada, Arizona, Florida, California were also most overvalued at their peak prior to the downturn. On the contrary, the 0-5% downturn states of North Dakota, Iowa, Texas, Kentucky, were all undervalued at their peak.
Valuation measured at varying peak dates in each region, correlated highly to the subsequent price downturns of varying duration in each region, which took about 4 years years on average post the peak to reach to bottom.
Pre-Global Financial Crisis: Overvalued U.S. States
Overvalued markets are hereby defined as ones having > 10% valuation. Overvalued U.S. states had median 26% valuation at their respective price peak. They experienced median 22% price drop peak-to-bottom.
Pre-Global Financial Crisis: Fairly Valued U.S. States
Fairly valued markets are hereby defined as ones having 0-10% valuation. Fairly valued U.S. states experienced median 11% price drop peak-to-bottom.
Pre-Global Financial Crisis: Undervalued U.S. States
Undervalued markets are hereby defined as ones having negative valuation. Undervalued U.S. states experienced median 4% price drop peak-to-bottom.
Below are the income drops by U.S. state post the Global Financial Crisis. Note the average income drop across all states was 4%. The figure seems in line with the median price drop peak-to-bottom experienced in undervalued U.S. states.
An interesting argument to be made is that as a whole, undervalued states did not drop in valuation terms, such as price/income ratio, instead prices followed the average drop in incomes only. In comparison, a very different dynamic was experienced in overvalued or even fairly valued states, which did drop in valuation terms.
Real Estate Market Valuation by U.S. County at Peak Pre-Global Financial Crisis
County-Level Regression at Peak Pre-Global Financial Crisis
At county level, valuation showed 73% correlation vs subsequent actual price drops, until bottom.
Intuitively, it may be harder to predict smaller geographies. While result is less compelling than the state-level regression, decent predictive power is present. Regression is performed on ~2700 U.S. counties using similar methodology and calculation.
Several concluding remarks to made:
Real Estate is a fundamental asset. Market valuation and downside risk may be measurable for fundamental assets.
Valuation measured at varying peak dates in U.S. states and counties, correlated highly to their price downturns post the Global Financial Crisis. Correlation was higher at the state level, than the county level.
Undervalued U.S. states experienced only a minor drop in prices post the Global Financial Crisis, in line with the drop in incomes, and as a whole didn't drop in valuation terms. Overvalued U.S. states experienced major drops in prices in line with their valuation.
Price-income ratios may shift 'regime' as population-housing supply ratio changes. The latter is observed in large desirable cities, where affordability may shift downward (price-income ratios shift upward) as housing shortage develops (population-housing supply ratio increases). Using a moving average for price-income ratios over a well chosen historical time window inherently may reflect affordability shifts due to housing shortage. A more complete study incorporating population and housing supply may serve to further increase predictive power.
Federal Reserve Bank of St. Louis, Federal Housing Finance Agency, Bureau of Economic Analysis
About the Author
Stefan Tsvetkov is the founder of RealtyQuant, a company that brings data-driven and quantitative techniques to the real estate industry. On a mission to add massive industry value through education, investment, technology, and analytics.
Financial engineer turned multifamily investor, analytics speaker, and live webinar host. He holds a Master's degree in Financial Engineering from Columbia University, and during his finance career managed ~ $90 billion derivatives portfolio jointly with colleagues.
Featured on multiple Podcast and Webinar events including InvestUp, Best Ever Real Estate Show, Discovering Multifamily etc. Organizer of Finance Meets Real Estate live webinar series.