Predicting stock prices with real estate data

By Gavriel Merkado| May 2018

REalyse investigates whether an alternative data model can be created for the real estate sector by focusing on the potential relationship between shares, assets and the market itself.

The alternative data market is growing. Investors, hedge funds and banks are always looking for new sources of data that can be used to make market predictions. In fact, prior to working in real estate, I created quantitative models and black box systems for this very use.

Occasionally, I still dabble in the markets to satisfy my own curiosity. Several months ago, it became apparent that there was a potential relationship between the price of listed real estate company shares, the value of the company’s real estate assets and, unsurprisingly, the real estate market itself.

I thought to myself, ‘Who else has better data on the real estate market than REalyse? And who would be better placed to determine what, if any, relationship exists?’

Getting it off the ground

I looked at established real estate companies, breaking it down by three segments — those building properties, those benefiting from the buying, selling and letting of properties, and those associated with the transaction process. The outcome was the below:

Building

• Barratt Developments (BDEV.L)
• Redrow (RDW.L)
• Taylor Wimpey (TW.L)

Benefactor of buying, selling and letting of property

Transactional associates

• Countrywide (CWD.L)
• LSL Property Services (LSL.L)
• Savills (SVS.L)

It became clear that some of these companies couldn’t be included in the research. Savills, for example, couldn’t because separating the influence of the UK market from international activities on its Profit and Loss statement was too involved for a top-level study. Countrywide would also have to be discounted, with its recent profit warnings appearing to be the result of poor management, rather than issues with the overall market.

When it came to selecting the data (1) for the study, I started with the £/sqft values from West London, Liverpool and South East London, and looked at the correlations between prices and returns.

The chart below shows the monthly correlations in price (not returns) of each of the different shares, and the postcode areas of South East London (SE), Liverpool (L) and West London (W).

For those using our Excel plugin, and want to replicate this themselves, use the code =REalyseSearch(“AvrgSqftSls(W,1995–02,2018–03)”), replacing ‘W’ with ‘L’ and ‘SE’ respectively.

Interestingly, Redrow and Barratt Developments show a 96% correlation. This makes perfect sense when you observe their share prices over the past 24 years:

Source: Google Finance

Likewise, the -53% correlation between Zoopla and LSL Property Services adds up as they have moved in opposite directions over the past few years.

As expected, Liverpool, West London and South East London all have high positive correlations — the highest being between South East London and West London, and a slightly lower (84–85%) correlation between London and Liverpool. So far, so good!

Flavour of the month

When plotting the correlation of month-on-month returns though, some of the relationships seen above fell apart. The positive relationships between some of the property developers and house builders, such as Redrow and Barratt Developments, stayed strong. But, surprisingly, the link between monthly changes in £/sqft values and share prices, or even other locations, dissipated — highlighting how monthly changes in each location don’t correlate with the monthly changes in stock prices.

Altering the lag between periods and changing the calculation to annual changes rather than monthly changes both had some impact in improving the correlations, but not enough to generate a worthwhile signal.

So far, the research had shown that there’s a strong positive relationship between the values of real estate and some share prices, but that link doesn’t initially appear to explain changes in price.

After closer investigation though, it became apparent that to plot share prices against real estate values, the data set would have to be divided into pre- and post-2008 — the data skewed by the volatility caused by the ending of a boom and the start of a bust in 2007/08.

The graphs below show how the relationship between stock prices and £/sqft values held relatively consistent both pre- and post-2008.

The past 114 months have shown a strong, positive, mean-reverting relationship between the development companies and the value of property.

Mean reverting?

Each time the ratio of the share price relative to the value of property exceeds a certain upper or lower boundary, the market tries to return those values to a more ‘normal’ level.

Looking at Barratt Developments and West London, for example, when the market went too far in either direction — as a result of share or property prices rising or falling too much (represented by the red circles) — the market would revert back towards the mean line (represented by the yellow circles) in the following month.

The proof is in the pudding

In an ideal world, the results above would be mixed with many other signals, and perhaps a range of different assets, locations and companies. But, for this study, a very simple signal can be created based on the relationship noticed above.

The below pattern is the outcome when calculating the ratio of Taylor Wimpey’s share price relative to the £/sqft price of properties in West London. A reliable signal can be detected by ‘buying’ TW.L shares when the ratio is below 1 standard deviation of its average (the green line) and ‘selling’ when the ratio is above 2 standard deviations of its average (the red line). In each case, the signal is generated at the end of one month, executed at the start of the next month and is held until the end of the month.

Over a 10-year model, the signal triggered 21 trades — 18 long, three short — of which 86% were directionally correct. The average gain was 11%, while the average loss was just -5%. The standard deviation of returns was 9% and the sortino ratio was 2.86, using an unrealistically high-yet-conservative risk-free rate of 5%.

The outcome was a 420% gain over a decade, which was roughly 17.9% compounded. This increased to 21.6% compounded if allowing the unutilised funds to be invested at the risk-free rate.

The above sounds impressive, until you realise that a buy-and-hold strategy of Taylor Wimpey shares delivered 25% compounded annual returns.

A whole new ball game?

The relationship certainly needs some more investigation. What happens with Taylor Wimpey may not be mirrored by Redrow, LSL Property Services and Zoopla, while different locations and different data sets (the number of transactions, time taken to sell or rent properties) could produce contrasting results.

From a first look though, it seems that it is possible to make predictions about the performance of real estate stocks using alternative data.

Please note: This shouldn’t be considered as investment advice, a recommendation or an indication to make any investment decision — it’s merely for information and curiosity purposes.

Sources (1) Monthly share price data was obtained from Yahoo Finance.

‘REalyse (Treex Ltd) does not provide any form of investment advice or property advice or any other regulated function. Note that any information or opinions, presented or referred to in this article are for information purposes only. Any actions taken by a reader are done entirely at their own discretion, you are responsible for your own investment decisions and hold Treex Ltd harmless from the results of any such decisions’. Whilst every effort has been made to ensure the accuracy of the information herein some inaccuracies may remain.’

Originally published at realyse.com.