RESOLVING THE HOUSE PRICE PUZZLE WITH ALTERNATIVE DATA

MacroXStudio
MacroXStudio

Key Takeaways:

·       The media, analysts, and academics, puzzled by the increase in US home prices in the face of soaring mortgage rates, have offered a wide variety of explanations

·       Neither of the two main stories – strong demand or constrained supply can fully explain US home price behavior

·       Using alternative data, MacroX shows that for 2018-2020 high demand was a driver, but 2021 onwards has been a purely supply-constraint story

·       The rich set of models have a 6-month lead on Case-Shiller, are intuitive, and scale internationally

·       Sample use cases include a better US CPI path for the next 1-2 years, a real-time evaluation of the Fed’s MBS purchases, and an understanding of the ongoing UK housing crash in real-time.

THE HOUSE PRICE CONUNDRUM: RISING RATES AND RISING PRICES

Over the last eighteen months, the Federal Reserve has raised interest rates at the fastest pace in history, which has led to mortgage rates almost doubling to 7% from early 2022. With home affordability falling dramatically – close to the 2008 financial crisis – many expected home prices to fall. Yahoo News predicted the biggest housing correction since WWII. In reality, home prices declined only modestly in H2, 2022, and since then the Case-Shiller index has been close to all-time highs.

Fig 1: Mortgage Rates have more than doubled in the past two years but house prices have steadily risen in 2023, rapidly approaching all-time highs

Inconclusive Academic Studies

Academics are divided on the effect of mortgage rates on home prices. Kuttner’s (2012) careful study of the US and several international markets via a combination of theoretical user cost models and empirical validation finds only a “modest effect” through various channels such as dynamic user cost, risk-taking, and credit.  For context, a simple user cost framework is noted below where the rent-to-price ratio is equated to the nominal long-term interest rate while considering property tax, tax savings, the risk premium, depreciation, and future price appreciation.

The rent-to-price ratio (R/P) is a function of nominal interest rates (i), property tax (T), tax savings (L), Risk Premium (RP), depreciation (Dep), and future price appreciation (Papp/P).

Williams (2015) concludes that home prices do react significantly to changes in interest rates but only with a lag of two years, while Case and Shiller (2003) find insignificant effects of mortgage rate changes on home prices.

The difficulty in separating “correlation and causation” is highlighted by almost all authors since interest rate changes do not occur in isolation but in conjunction with many crucial macroeconomic variables. For example, late in the cycle, home prices may be increasing as the Federal Reserve hikes interest rates to combat an “overheating economy’, and since monetary policy acts with a lag, interest rates may still be increasing but home prices start declining towards the end of the cycle. From a practical standpoint, most theoretical “value” models (such as equation 1) tend not to be helpful over a 6-month horizon as little changes.

Many Stories

The seemingly counter-intuitive nature of home price increases has resulted in widespread attention, and many explanations from the media, analysts, and academics have emerged. There are two major strands – constrained supply and an unexpected rise in demand.

Strong Demand

Some recent explanations highlight the role of demand-based factors ranging from wealth built up during the pandemic to a shift in preferences to working from home. Some studies suggest that 50% of the 23.8% increase in national house prices (from 2019 to 2022) can be explained by the shift to remote work. A recent Economist article asserts how strong the housing demand is.

Fig 2: Home Affordability is at an all-time low (Atlanta Fed) which bodes poorly for the strong demand story in recent times.

The strong demand story quickly falls apart in recent times with home affordability at an all-time low as Figure 2 shows. Additionally, most banks have been tightening lending standards from July 2021 onwards (Figure 3) which hurts housing demand.

Fig 3: Most banks have been tightening loan standards post July 2021. Source – Tradingeconomics.com.

Constrained Supply Stories

The other major branch of explanation is around constrained supply. The recent mortgage rate increases make changing homes – which implies refinancing at significantly higher rates – an unattractive proposition. Figure 4 approximates the difference in recent vs. historical mortgage rates for an existing homeowner who typically stays in a home for 10-12 years.

Fig 4: The Mortgage Funding Differential – Current Rate vs. Old Financing Rate is high.

Recent media coverage has highlighted decreased mobility and a recent poll by Zillow found rate-locked homeowners to be nearly twice as unlikely to consider selling. A sharp decline in existing home sales that form the bulk of housing transactions as compared to new home sales – which are at neutral levels – lends further weight to the “constrained supply” story.

Fig 5: Existing Home Sales have fallen much further from historic norms compared to New Home Sales

However, given the variation in home availability – such as the spike higher in late 2020 and early 2021, it is not credible as the sole explanation for the entire period.

MACROX’S FRAMEWORK TO SYNTHESIZE AND QUANTIFY DEMAND AND SUPPLY STORIES

At MacroX, we follow three major steps – we broaden the data sources to non-traditional ones, separate them into demand and supply factors, and after quantifying each factor synthesize it into a comprehensive model.

The resulting model is intuitive since it builds upon the insights present in the various stories, highly predictive as it predicts Case-Shiller 6 months in advance of its publication, and scalable as it translates internationally. We describe the steps in detail below.

(i)                  Using Alternative Data

As a reminder, alternative data refers to data gathered from non-traditional data sources. This can refer to credit card transactions, search data, satellite data, and many others. Some housing analysts already use the data provided by real-estate companies, but MacroX supplements this with other, complementary alternative data – including Satellite, News, X(formerly Twitter), Search, transactions data, etc. For example, it is plausible that social media data could provide valuable information on the future intentions of both buyers and sellers of a house which can be validated by consumer spending data.

Fig 6:  Relating social media topics to the user journey helps construct portfolios. Hundreds of such sample paths are constructed with relevant localization.

(ii)                Identifying and Quantifying Demand and Supply Separately

It is crucial to separate and quantify demand and supply factors separately for cleaner identification. For instance, the social media behavior of a person looking to list their home for sale, or one looking to buy properties may be subtly different. Using the alternative data sources described above, we construct hundreds of ‘portfolios’ that explain the typical behavior of users in that location. The implementation of the same concept can be different in different regions according to the regulations and customs – some cities are very strict about building permits and some are quite liberal.

Fig 7: Our measure of housing demand combines forward-looking social data with accurate broker and other micro data. It currently shows an uptick in demand from low levels.
Fig 8: Our measure of housing supply forward-looking social data with accurate broker and other microdata. The recent supply has remained low.

(iii)               Synthesizing Into a Model

While there are many ways to combine the rich alternative-data-based feature demand and supply feature sets, we illustrate a simple methodology. First, we construct a “meta-portfolio” of all demand and supply-based features and then we subtract the supply meta-portfolio from the demand meta-portfolio to generate an “excess demand” meta-portfolio. A visual inspection of Figure 9 reveals that the golden line – the MacroX excess housing demand portfolio (golden) – leads the Case-Shiller index (white) by a few months.

Fig 9: Our measure of excess demand tracks house prices well – after a period of extreme excess demand, housing supply and demand are currently well-balanced

Fig 9: Our measure of excess demand tracks house prices well – after a period of extreme excess demand, housing supply and demand are currently well-balanced

Model Statistical Tests

Following the visual examination, we conduct statistical tests.

Initial Regressions

Among the two main indices – Case-Shiller and Zillow, Case-Shiller leads Zillow, (confirmed by Granger Causality tests) so we focus on that. But results for Zillow are similar. We find that the MacroX excess demand measure leads the Case-Shiller index by 4 months in calendar time – meaning at month t, we can predict the month t+4. Additionally, in real-time the Case-Shiller index is released with a 2-month lag, so it can combine to an effective lead of 6 months. Note that we do control for the illiquidity of Case-Shiller by including prior lags.

Figure 10: The Excess Demand Measure retains significant predictive power over Case-Shiller even 3 months ahead controlling for Case-Shiller return today. The Newey West and Robust standard errors are similar.

Robustness Checks for the Illiquid Case-Shiller

Unfortunately, MoM changes in Case-Shiller are highly autoregressive due to the illiquidity of the housing market with a simple AR(2) regression explaining ~85% of the variation of the index. Both levels and changes in excess demand – notably constructed as a weakly stationary process with finite variance and long-run mean of 0 – are crucial in determining the effects on house prices. Given the illiquidity of Case-Shiller, we perform further verifications in addition to using its own lags.  

Using non-overlapping quarterly growth in the index provides a way of removing much of the autoregressive nature of the index. The QoQ index may be relevant not just due to its statistical properties, but also in corresponding more closely to a slower real-estate and related securities investing cycle.

When we add the change in our measure of excess demand in the quarter to the above regression, the variation in the QoQ change in the Case-Shiller explained almost doubles to ~35% from 19.5% and the coefficient is statistically significant at the 5% level:

Figure 11: MacroX’s measure of excess demand doubles the variation explained for house price growth. The coefficient is both economically and statistically significant

Furthermore, our measure of excess demand is highly economically significant, as a 1 standard deviation move higher is associated with a 1 standard deviation move rise in prices over the quarter. Therefore, using our measure of demand-supply can vastly increase the ability to predict current quarter growth in house prices (particularly relevant as the Case-Shiller price index is only published with a two-month lag!).

We also test to see whether our measure of excess demand Granger causes changes in house prices as measured by the Case-Shiller Index and the Zillow Home Value Index. In both cases, we find that this is indeed the case with 2 lags of our measure being the optimal amount.

Figure 12: Granger Causality Tests corroborate the findings

MACROX REAL ESTATE MODEL SAMPLE USE CASES

 A.      Explaining the US Housing Market – Supply and Demand Are Important at Different Times

Our excess demand measure successfully explains the Case-Shiller price variation as noted above. However, to relate it to the two major demand and supply stories, we perform another set of tests. Specifically, we disaggregate the measure into supply and demand variables and regress both on three time periods. The first is from 2018 to 2020, and the second is from 2021 to today.

The betas and t stats of the regression over various sample periods reveal the rich dynamics. In Figure 13 the height of the teal bar shows the importance of the (standardized) demand variable as indicated by the t-statistic, and the importance of the (standardized) supply variable is indicated by the height of the red bar which again relates to the supply t-statistic. The graph reveals:

–          For 2018-2020, high housing demand and constrained supply both matter (as measured by t-stat). In other results (not included here), we find that demand is more important in the next month and supply is not.

–          For 2021-today, constrained supply is the most important factor and demand does not matter significantly in this month or the next.

The model indicates that anyone looking for a housing crash over the next few months will be disappointed as demand is increasing and the supply remains low.

Fig 13: This figure shows a “horse race” of supply (red) and demand (teal) factors in explaining Case-Shiller returns with the height of the standardized coefficients corresponding to the t-statistic. We find that higher demand was important in 2018-2020 but not in 2021-today. Supply shortages are the main factor in the 2021- today period.

B.      Predicting US Rents and Inflation

Following a similar methodology, we construct the “Rent Excess Demand” measure and predict the US rent trajectory over the next 3-6 months. This is not only useful for consumers, but for real estate investors and for predicting the path of inflation in the coming months.

Fig 14: A similar approach provides good results for rents as well – an increase in rental supply has caused rental growth to be muted in 2023

Rent is a large part of various inflation baskets but owing to the convoluted methodology acts with a 12–15-month lag on the government measures. Having the 3 to 6-month rent path forward adds to our ability to nowcast inflation’s trajectory even further into the future. Here is our detailed piece.

 Fig 15: We find the greatest correlation between rents and CPI OER when we lag rents by 12- 15 months – see more

C.      Going International – the UK Housing Downturn

A good test for any model is to see if it can explain another country. When we apply the model to the UK housing market, we can see that the excess supply coming online in the UK is much higher than the lower demand, putting more downward pressure on house prices as compared to the US.

Fig 16: A different story for the UK with high supply and neutral demand leading to falling house prices for much of 2023

D.      Policy Use a.k.a. The Fed Might Have Inflated the Bubble

Since housing is one of the largest components of wealth for most individuals in advanced economies (Muellbauer et al.) and is funded by a substantial financial sector, it is of concern to central banks. A policymaker might face a tradeoff between the housing sector prospering vs. the overall economy being stable (John Williams (2015).

Whether home prices are increasing due to changes in demand or supply, might have profound policy implications – the demand-driven changes are potentially more in the purview of the central bank to influence via interest rates, whereas the supply-based changes might have broader policy responses.

For example, the Fed’s MBS purchase strategy during the pandemic may have benefitted from the nuance provided by clear identification. In March 2020, we can see that housing demand collapsed and thus the Fed’s purchasing of MBS made perfect sense to support the housing market. However, the Fed may have been less keen to engage in MBS purchases in 2021 if it could clearly see that housing demand was elevated relative to recent averages. The Fed’s unnecessary support to the housing market might have contributed to the “affordability crisis.”

Fig 17: The Fed might not have continued buying MBS in 2021 if it could see the recovery in housing demand from the pandemic shock in real time.
Fig 18: City-level analysis of work, demographics, and other patterns along with demand-supply dynamics could be a powerful new tool for investors looking for a 5-year lead.

CONCLUSION

We find that an alternative data-based framework that identifies and quantifies various supply and demand narratives into a rich series of models successfully explains the rise in US home prices over the last few years – with both supply and demand factors being important at different times. The intuitive model is quite powerful and can nowcast the Case-Shiller returns 6 months in advance, the US rent component of CPI about 1-2 years in advance, and even scales to other international markets such as the UK quite successfully. 

The next major iteration involves overlaying the model with rich microeconomic data such as transactions, demographics, and local and remote employment patterns to deepen its application.

You may also like

Sign up for our newsletter