<<< Back

Methodology - RCIRE (U.S.) Index

The Rational Consistency Index for U.S. Real Estate (RCIRE) is a machine-learning-based dynamic metric designed to quantify how rationally the U.S. housing market interprets news. Unlike traditional sentiment indicators, RCIRE does not measure emotions in news article itself. Instead, it measures how much emotional framing affects the market’s implied response, and how relatively rational the current real-estate market is, as learned from historical data.

“We help policymakers, researchers, and market participants monitor housing-market phases, identify periods of emotional amplification, and signal moments of potential irrational exuberance."

--Thomas Chan (Founder of RECIRE index)

Rationale Behind

RCIRE is built on a simple idea: a purely rational market should react the same way to a “original news article” and to “its emotion-neutral version”, as long as the underlying information is identical. By using AI and machine learning to learn from both versions of the news and compare their explanatory power for current real-estate dynamics, RCIRE quantifies how consistent real estate market reactions to the “original” and “neutralized” versions of the same news are, as a proxy of current market’s degree of rationality. The closer the index is to one, the more rational the U.S. real estate market behaves; the larger the gap, the stronger the influence of emotional framing.

News 1

" Texas home prices soared unexpectedly as buyers rushed back into the market, fueling fears of renewed overheating."

News 2

"Taxas home prices increased as transaction volume rose, reflecting higher buyer participation."

Shall yield no different response for a purely rational market participant.

Brief Intro:

1. News Collection and Text Embedding

Real-estate-relevant news is collected and transformed into high-dimensional vector representations using large large model (LLM) embeddings.
These embeddings capture not only semantic meaning but also the affective tone, information density, uncertainty, and narrative structure of each article.

2. Market-Calibrated Return Prediction Function

A machine-learning function is trained to predict future real-estate market movements—such as REIT returns or housing index variation—from news embeddings:

Because the model is calibrated on realized market data, f(·) becomes a data-driven approximation of the market’s semantic response function — capturing how the market has historically translated textual information into price movements.

3. Affective Neutralization of News

TextFor each news item, a parallel emotion-neutral version is generated.

The goal is to remove emotional phrasing while preserving factual structure and informational content:

Example of Neutralization

Original:

"Texas home prices soared unexpectedly as buyers rushed back into the market, fueling fears of renewed overheating."

Neutralized:

"Taxas home prices increased as transaction volume rose, reflecting higher buyer participation."

What changed?

“soared unexpectedly” → “increased”
“rushed back into the market” → “transaction volume rose”
“fears of renewed overheating” removed entirely (emotion only)

What did not change?

The underlying information: prices rose, demand increased.This controlled transformation allows RCIRE to isolate the specific effect of emotional framing on market responses.

4. Differential Sensitivity to Emotional Framing

RCIRE compares how the market-trained ML model responds to the original vs. neutralized news:

Small Δ → emotional wording has little influence → market behaves rationally
Large Δ → emotional framing materially alters expected returns → market displays behavioral sensitivity

This differential response is the backbone of the Rational Consistency Index.

5. Rational Consistency Index (Unsigned)