The Sell Odds Methodology

A listing that fails to sell is evidence to investigate, not proof of one cause. Price is one part of that investigation, alongside condition, presentation, documentation and timing. I built Sell Odds so the price is never set blind.

Why is the San Gabriel Valley hard to price?

Because it is not one market, and nothing in it behaves the way an automated estimate says it should. Pasadena, South Pasadena, Altadena, Sierra Madre, San Marino, Alhambra: none of these is one market. A Craftsman in Bungalow Heaven and a Craftsman in Altadena share an architectural style and almost nothing else. A historic-district designation moves the price in one direction. The Mills Act moves it in another. Lot configuration, school boundary, foothill aspect: each one is a number that moves the price, and none of them is what a guesstimate sees.

What question does Sell Odds actually answer?

One question, and it is not what the house is worth. Sell Odds is an empirical probability engine, and it asks whether this property will actually sell at this price. Worth is one number. Whether a property will sell at a given price, measured against the record of what this market has already done, is a different number entirely, and it is the one that decides whether you move or you sit. Sell Odds is built to measure the second number. That is the only question it asks, and this page is the record of how it answers.

Sell Odds began as its own product, separate from any single market. Its proving ground and full development story live at sellodds.com. What runs on Arroyo Casa is the same engine, calibrated to the cities I serve in the San Gabriel Valley as a REALTOR. Same math. Same data discipline. To see it work, find any property in the Arroyo Casa search, click Get Probability Score, and watch the Crystal Ball run Sell Odds live against that property, at that moment. The number it returns is the market’s answer. The engine answers exactly one question; the questions that surround it, when to list, what to prepare, what the city requires before escrow can close, are gathered in the Seller FAQ: Historic and Architectural Homes in Pasadena and the San Gabriel Valley.

Where does the math behind Sell Odds come from?

It comes from the discipline finance has used for sixty years. In the early 1960s, a mathematics professor named Edward Thorp published Beat the Dealer and proved the blackjack table could be beaten with probability math. Then he took the same math to Wall Street. The fund he co-founded in 1969, Princeton Newport Partners, is widely regarded as the first quantitative hedge fund, and its record still reads like a misprint: nineteen winning years in nineteen. Empirical math built modern quantitative finance on that foundation. It is how serious money has measured probability ever since: not by opinion, but by the record of outcomes.

Why has residential real estate not used this math?

It never adopted it. Six decades later, walk into any residential transaction and look for that math. It is not there. What consumers get instead is the Zestimate. Zillow, Redfin, homes.com, all the portals people doomscroll in bed at half past midnight: every one of them is a lead-generation business, and the estimate is the bait. Your contact information is the product. Those platforms want the listing. They are not built to help you make a better-informed decision, and not one of them can tell you whether your property will sell at the number on the screen.

An investment-grade discipline for measuring probability sat in plain sight for sixty years while the industry handling the largest transaction of most families’ lives ignored it. I couldn’t watch it anymore. So I designed what did not exist.

What data does Sell Odds run on?

Sell Odds runs on a live connection to the California Regional Multiple Listing Service. The data is not scraped, not estimated, not algorithmically generated. The engine queries it live for every score, and every score is calculated fresh, for the property in question, at the moment it is requested. No stale cache. No pre-computed estimate.

Why are active and pending listings excluded?

Because neither one proves anything about market acceptance. The engine uses sold and unsold outcomes only. Active listings are excluded, because a price sitting on the market proves nothing about market acceptance. Pending listings are excluded, because they have not closed. What remains is the only empirical record of what the market did and did not pay: the properties that closed, and the properties that came to market, sat, and walked away unsold.

Here is the part that matters. Every licensed agent in California can open the MLS and look at that failure record, one listing at a time. Almost none of them can do anything with it. Nothing in an agent’s toolkit aggregates tens of thousands of outcomes and turns them into a probability, and no human reads that many outcomes off a screen. Software has to do it. The software did not exist. So I built it. That is the whole point.

How is a Sell Odds read different from a CMA?

It is held to a different standard of evidence. Sell Odds does not take its method from the appraisal world. Its method is empirical math, the same discipline finance runs on. What it shares with appraisal is the standard of rigor. Licensed appraisers are trained, tested, and bound to strict, recognized approaches to value: the sales comparison approach, the cost and replacement approach, the income approach, and the disciplined variants built on them. The approach changes with the property. The discipline never does: defined criteria, defensible adjustments, a conclusion that can be audited.

Real estate agents carry no such requirement. Nothing in an agent’s licensing educates them in a valuation discipline, and nothing obligates them to acquire one. So what most sellers receive as a CMA, a Comparative Market Analysis assembled at the kitchen table on the day an agent is trying to win the listing, is photographs of nearby homes with similar bedroom counts, presented as analysis.

And there is a harder truth about that kitchen-table package. The comps in it are routinely cherry-picked to flatter the number, because the agent’s goal that day is your signature, not your outcome. The industry has a name for this. It is called buying the listing. It is one of the oldest open secrets in the business, it has run for decades, and the overpricing born at that table is where failed listings begin.

A conventional market analysis and a Sell Odds read provide complementary evidence. The Probabilistic Marketing Analysis shows how the estimated probability of sale changes with the asking price. Empirical math, held to appraiser-grade rigor, applied at a scale no individual can work at. Open Comps and Analysis on any Arroyo Casa property and you see the sold comparables the engine actually used, each carrying a confidence score for how closely it matches the subject on square footage, bedrooms, bathrooms, and structural type, with a compare view showing exactly where it differs. Sold data, appraiser-grade, on demand, free.

Eight Filtering Gates. Zero Assumptions.

Every score runs through eight gates before the engine produces a single number.

Adaptive tiered radius. The comparable search starts tight around the property and widens only when the data requires it. A hard geographic ceiling prevents cross-submarket contamination.

Structural matching. Bedroom count, bathroom count, and property type are gated, not suggested. A four-bedroom home is never compared to a two-bedroom.

Three-year rolling window. Old enough to capture full market cycles. Recent enough to reflect current conditions.

Price banding. Properties far outside the realistic range are excluded before they can distort the measurement.

Sold and failed outcomes. The engine reads what sold and what did not: expired, withdrawn, canceled. Both sides of the record, always.

Minimum thresholds. If the data is not sufficient for a reliable measurement, the system says so. It does not fabricate confidence.

Lease contamination filtering. Rental listings in the MLS are detected and excluded. A rental comp is not a sale comp.

Property subtype matching. Single-family to single-family. Condominiums to condominiums. Manufactured to manufactured. No blending.

Is the Sell Odds number a prediction or a measurement?

A measurement. The probability Sell Odds returns is a direct empirical reading: out of comparable properties listed within this price range, the percentage that successfully closed versus the percentage that failed. That is empirical math at work. The probability is a count, not a guess. A high number means the market has a strong track record of closing at that level. A low number means properties in that range have a history of sitting, expiring, or being pulled.

There is no black box behind the number. No machine learning model producing an output nobody can explain. The Crystal Ball does not tell you what to do. It shows you what the market has done.

What does moving the price slider show?

It shows the shape of the curve, which is where the leverage lives. A single probability at a single price is useful. Move the price slider on any property and the Crystal Ball recalculates across 201 individual price points, so you can see exactly where the probability of sale begins to drop, and how sharply. Most consumers have never seen what happens to a property’s odds when it is overpriced by $10,000 or $20,000. Now they can.

What can the Sell Odds engine not see?

Condition, and I will not pretend it can. A home two doors from yours sold last spring after a $140,000 renovation. Yours has the original kitchen. The MLS has no required field for that. Renovation detail lives, when it lives anywhere, in free-text public remarks, and agents are notoriously inconsistent with remarks. I spent months trying to build a feature matrix that could reliably read remodels, additions, and ADUs out of public remarks. The remarks are too inconsistent. Rather than let the engine guess, I held it to the same standard as everything else on this page: it does not fabricate confidence.

I built and remodeled homes for twenty years. I assess the house in person, including its condition, originality and work that may need specialist investigation. That assessment informs my pricing or offer advice alongside the probability estimate. The engine measures sold and unsold outcomes; I assess the house. What I am reading when I do it is set out in Owning the Architectural Home and in Buying an Architectural Home.

Why is the public Crystal Ball number citywide only?

Because the hyperlocal record cannot lawfully be published to the general public. The probability the public Crystal Ball shows is the citywide layer. So is the Market Leverage reading: supply pressure, sell-through rate, price realization, and days on market, combined into a single reading of who holds the advantage at the table. All of it runs free, on any property, on demand.

The hyperlocal layer is a different instrument. A heat index calibrated to the immediate radius around one address, at a tighter standard of criteria than the citywide view. Failed-comp forensics on the properties directly comparable to yours. A pricing pressure curve for your street, not your city. And the failure record itself: expired, withdrawn, and canceled listing detail is governed under IDX and VOW regulations and cannot be displayed to the general public on any website, this one included. Every licensed agent has access to that record. Almost none can turn it into a measurement.

That layer I deliver at your table. And when I do, I unlock the record live: the unsold addresses, the failed listings, on the screen, for your property at the center of the conversation. That is the quality of proof I bring.

So if you ran your property through the Crystal Ball and the number came back lower than you expected, do not price off that number, and do not dismiss it either. It is the citywide truth. The delta between the citywide reading and the hyperlocal one is almost always significant, and that delta is the entire conversation. Reach out and we will run it together: request your Sell Odds read and in-home analysis.

Is Sell Odds an appraisal?

No. Sell Odds is not an appraisal, not a guarantee of sale price, sale timing, or marketability, and not investment, legal, or tax advice, as the disclosure at the top of this page sets out in full. Past outcomes do not predict future results, and property-specific factors, market conditions, timing, presentation, and other variables outside the model’s scope can move actual outcomes materially.

Where Sell Odds output informs a list-price or offer-price recommendation for a specific property in the course of brokerage services, that recommendation is a broker opinion of value provided in the ordinary course of California broker brokerage activity. It identifies its market-data basis, it names this page as the published basis, and it is not an appraisal under California Business and Professions Code §11302(b).

I am Greg Anderson, a California REALTOR since 1990, and Arroyo Casa is my practice across the architectural and character homes of the San Gabriel Valley. The engine is one instrument in that work and it is published here so you can inspect it. What it measures and what I assess are two different things, and you are entitled to both before you set a number. When you want them together on your own property, Selling an Architectural Home is where that conversation starts.