It must have been quite a discovery for
Francis Galton and his scientific peers in the 19th century to acknowledge the
popularizing of the new regression toward the mean. The humble finding was
founded in methodology when studying the predictions of height within families.
Like the Elixir of Life, science had a successful stab at prediction. A brave
new world was paved ahead in all fields of science, including asset prices of
real estate. With the methodology known, it took a near century for real estate
data to be accessible.
With government monopoly on transaction data
and the need for efficient real estate taxation starting in the 1980s,
Automated Valuation Models (AVMs), the regression was starting to be utilized
en masse. The past half century was an evolution of data from a monopolistic
framework to a situation of monopolistic competition, enriched and combined
with Artificial Intelligence. Today we stand at a new brave new universe of
asset pricing. It is easy to race down the bandwagon, but as Francis Galton and
his peers later learned and history sadly pivoted toward the extremely
dangerous pitfall; this article peels some of the onion rings on the historical
Kafka lessons learned on regression, data, and humanity’s role.
This article focuses on data versus subject
matter experts, regression methodologies, and the advent of new technologies
promising instead a brave new universe, but questions if these technologies
stand on the shoulders of giants or hallucinating phantoms.
The folly, Pi, and the designers of AVMs
Often overlooked, the best academia brings
together forces of two different disciplines. In such, the world-famous asset
pricing model Black-Scholes (1973) for pricing options is such an example.
Myron Scholes was the economist, but it was Fischer Black, with a background in
thermodynamics, that brought the key ingredient ‘sigma’ (volatility), as in the
rise and fall of temperature, that made the model work. It is precisely the
best of both worlds that often lacks with the AVMs available to the market. The
pitiful, like the later destructive application of Francis Galton, AVMs are
frequently developed by data scientists. Data scientists, however fabulous with
their machines, but as McKinsey Quarterly in 2001, the popularized the
Pi-shaped approach. Instead, the AVM tech is shaped by T-shaped people. The
Quarterly reported clearly the risks. Data scientists traditionally are the
former, and it should be worrying that they instead, or lack conjunction with,
trained real estate professionals, ‘predict’ prices. Lesson one: when vetting
AVMs, do not review the back tests; instead break down the team experience and
identify the Pi-shaped AVM organizations.
The data normalization trap in real estate
As most elementary children know, one needs
two points to draw a line. But what if those coordinates are misaligned and
that line goes to infinity? For sure the Mars shot is missed. That is why in
real estate, value adjustments are dirty and gritty, time-consuming work.
Validating each point and adjusting so the stars are aligned is precisely the
human work of a real estate professional. One can pour a bucket into a funnel
and allow the marbles/data points to fall into a normal distribution, but
little use it is when the very axiom is incorrect. And incorrect is frequently
the norm; it is here the hybrid form of AVM is more powerful, together with
alternative data. Using house prices to derive, with a regression, a prediction
of price is folly. The prices must be adjusted to values and then enriched with
alternative data that power not a pricing model, but a valuation model. Most
valuation models are driven by price; a logic error so basic as Watson
frequently hears. Lesson two: many tech companies fall short utilizing
misaligned pricing data to engine their models.
The Cannery, AI, and Pied the Piper have a
chat
Every generation has a magic wand; in this
time and age, it is Artificial Intelligence (AI). For sure, it is a marvel of
the day and age that brings society nearer to a new Brave New Universe. In the
AI revolution, a word is missing. That word is a common thread in this paper;
it is HI. The HI stands for Human Intelligence that has been around since the
dawn of time. Markets flip, storms come, a canary in a coal mine knows it, yet
AI dreams in a virtual reality world. The AI agent, trained with the best of HI
words in poetic prose, enacts the 13th-century myth of ‘The Pied Piper of
Hamelin.’ It is a tale that teaches about trust, responsibility, and
consequences; do historians not always say ‘history repeats itself’.
As any good engineer will ask first, what are
the redundancy systems in place? It is curious how lacking that question is
with AVMs specifically driven by AI. It is very worrisome when knowing that
real estate debt (mortgage) accounts for between forty and sixty percent of
Gross Domestic Product for many EU countries. It is time for HI to ask critical
questions to the governing bodies, system banks, and tax collectors mandating
that extra safeguards are in place. Fans of ‘Little Britain’ know all too well
the pun ‘Computer says NO’. An engineer in kindergarten terms can explain the
term Catastrophic Failure, or Myron Scholes in his work on Risk Management and
Fischer Black in thermodynamics the concept of Pressure Drop; the writing is on
the wall. Lesson three: the iron triangle scope is being changed with AI; what
roles safeguard HI in the decision-making process?
The final tune and historical woes of
regression
Real estate has a huge contributory value to a
country’s GDP; unexpected swings in GDP bring market uncertainty, economic
destabilization, and the polls react. One asset class more than any other asset
class, real estate price swings like the Sword of Damocles continuously above a
nation’s well-being. The real estate valuation models used to predict prices
are more often AI models trained by T-trained people using unvalidated dots
with no canaries in a cage, and governing bodies are rewriting the notes to Pied
the Piper tunes. This is no brave new universe, but a world of folly that
awaits that Kafka knew all too well.
