The Sell Odds Methodology

Most listings that fail in the San Gabriel Valley did not fail because the property could not sell. They failed because a price was set without the data that would have said what was likely to happen. I built Sell Odds so that call never has to be made blind again.

The valley makes that call harder than almost anywhere. Pasadena, South Pasadena, Altadena, Sierra Madre, San Marino, Alhambra: none of these is one market, and none of them behaves the way an automated estimate says it should. 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.

The Single Question

Sell Odds is an empirical probability engine, and it asks one question: will this property actually sell at this price? Not what it is worth. 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.

The Math Real Estate Ignored

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.

Real estate 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.

The Record Nobody Shows You

Sell Odds runs on a live connection to the California Regional Multiple Listing Service, the largest MLS in the United States. 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.

It 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.

An Appraiser’s Rigor. A Quant’s Math.

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.

Sell Odds replaces that package with a PMA: a Probabilistic Marketing Analysis. Built from data, not photos. 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.

Not an Algorithm. 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.

Price Elasticity, Made Visible

A single probability at a single price is useful. The shape of the curve around it is where the leverage lives. 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.

The Layer the Engine Cannot See

There is one thing the engine cannot see, and I will not pretend it can: condition. 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.

So the quality-tier adjustment is done by me, in person. I built custom homes for twenty years, and that background is exactly what this layer requires: I know what a $140,000 renovation or an $80,000 kitchen adds against the comp down the street, because the photographs show us what that comp did and did not have. The engine measures the market. I measure the house.

Citywide Is Free. Hyperlocal Is in Person.

One more thing about the number on your screen, because it frames everything: 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.