Rental Desk: Live Scoreboard
Two teams of AI agents each picked 25 US cities to own as rentals, using only Zillow home-value and rent data. Their portfolios are on paper (no money is invested), and this page grades them every month against the obvious alternatives. The backstory, including how the agents argued their way to these rules, is on Teams of AI Agents.
Status: 0 months since the books were bought on September 30, 2026. Latest Zillow data: September 2026. The first graded month arrives in late November, after Zillow’s mid-month release.
The contestants
- Desk v2 (season 2, locked October 2026): the 25 highest gross-yield cities that have at least two years of rent history, at most 3 per metro. After charging each city its real property tax, insurance, upkeep, and buying and selling costs, no cleverer rule beat this in the agents’ backtests.
- Desk v1 (season 1): yield plus 12-month price momentum, with filters. It beat the naive screen in every backtest, then lost to it from 2023 to 2026. It’s still held, to see whether that repeats.
- Naive top-25 yield: the 25 highest gross-yield cities, no filters. This is the bar to beat.
- Strong Buy tier: every city my US Rental Investment Markets page labels Strong Buy (price-to-rent under 15), equal-weighted.
- All eligible cities: the average city, for scale.
Standings
| Basis | Portfolio | Return since entry |
|---|---|---|
| Gross (50% rule) | Desk v2 (gated yield) | +0.0% |
| Gross (50% rule) | Desk v1 (yield + momentum) | +0.0% |
| Gross (50% rule) | Naive top-25 yield | +0.0% |
| Gross (50% rule) | Strong Buy tier | +0.0% |
| Gross (50% rule) | All eligible cities | +0.0% |
| Net of real costs | Naive top-25 yield | -8.1% |
| Net of real costs | Strong Buy tier | -8.1% |
| Net of real costs | Desk v2 (gated yield) | -8.2% |
“Gross” is price appreciation plus half of rent collected (the 50% rule). “Net” charges each city its own costs and assumes a sale at the end of the month, so it starts about 8% in the hole from buying and selling costs and has to earn its way back.
The 2023 book, still running
Season 1’s locked 2023 picks are graded on the same terms and keep compounding. Three years in, they were up about 23% on the 50% rule.
What each desk owns
The two desks share 2 of 25 cities.
Desk v2
| # | City | Metro |
|---|---|---|
| 1 | Century Village, FL | Miami-Fort Lauderdale-Pompano Beach, FL |
| 2 | Glasgow Village, MO | St. Louis, MO-IL |
| 3 | Gary, IN | Chicago-Naperville-Elgin, IL-IN-WI |
| 4 | Lauderdale Lakes, FL | Miami-Fort Lauderdale-Pompano Beach, FL |
| 5 | Jackson, MS | Jackson, MS |
| 6 | Bellefontaine Neighbors, MO | St. Louis, MO-IL |
| 7 | Ojus, FL | Miami-Fort Lauderdale-Pompano Beach, FL |
| 8 | Camden, NJ | Philadelphia-Camden-Wilmington, PA-NJ-DE-MD |
| 9 | Ferguson, MO | St. Louis, MO-IL |
| 10 | Cleveland, OH | Cleveland-Elyria, OH |
| 11 | Maple Heights, OH | Cleveland-Elyria, OH |
| 12 | Inkster, MI | Detroit-Warren-Dearborn, MI |
| 13 | McKeesport, PA | Pittsburgh, PA |
| 14 | Bessemer, AL | Birmingham-Hoover, AL |
| 15 | Garfield Heights, OH | Cleveland-Elyria, OH |
| 16 | Decatur, IL | Decatur, IL |
| 17 | Pontiac, MI | Detroit-Warren-Dearborn, MI |
| 18 | Calumet City, IL | Chicago-Naperville-Elgin, IL-IN-WI |
| 19 | Hazel Park, MI | Detroit-Warren-Dearborn, MI |
| 20 | Birmingham, AL | Birmingham-Hoover, AL |
| 21 | Binghamton, NY | Binghamton, NY |
| 22 | Port Arthur, TX | Beaumont-Port Arthur, TX |
| 23 | Baltimore, MD | Baltimore-Columbia-Towson, MD |
| 24 | Peoria, IL | Peoria, IL |
| 25 | Dayton, OH | Dayton-Kettering, OH |
Desk v1
| # | City | Metro |
|---|---|---|
| 1 | Abilene, TX | Abilene, TX |
| 2 | Southampton, NY | New York-Newark-Jersey City, NY-NJ-PA |
| 3 | Pennsauken, NJ | Philadelphia-Camden-Wilmington, PA-NJ-DE-MD |
| 4 | Rockford, IL | Rockford, IL |
| 5 | Lawton, OK | Lawton, OK |
| 6 | Toledo, OH | Toledo, OH |
| 7 | Peoria, IL | Peoria, IL |
| 8 | Sag Harbor, NY | New York-Newark-Jersey City, NY-NJ-PA |
| 9 | Springfield, IL | Springfield, IL |
| 10 | Beloit, WI | Janesville-Beloit, WI |
| 11 | Waukegan, IL | Chicago-Naperville-Elgin, IL-IN-WI |
| 12 | Utica, NY | Utica-Rome, NY |
| 13 | Fort Lee, NJ | New York-Newark-Jersey City, NY-NJ-PA |
| 14 | Jacksonville, AR | Little Rock-North Little Rock-Conway, AR |
| 15 | South Bend, IN | South Bend-Mishawaka, IN-MI |
| 16 | Binghamton, NY | Binghamton, NY |
| 17 | Saginaw, MI | Saginaw, MI |
| 18 | Alton, IL | St. Louis, MO-IL |
| 19 | Bay City, MI | Bay City, MI |
| 20 | Syracuse, NY | Syracuse, NY |
| 21 | Bartlesville, OK | Tulsa, OK |
| 22 | Tupelo, MS | Tupelo, MS |
| 23 | Hamden, CT | New Haven-Milford, CT |
| 24 | Berkeley, MO | St. Louis, MO-IL |
| 25 | Phenix City, AL | Columbus, GA-AL |
Known weak spots (not overridden)
The rules are systematic, so odd picks stay in rather than getting hand-removed. Watch these:
- Century Village, FL (v2’s top pick) is a 55+ condo community, where rental restrictions are common. Its yield may not be achievable.
- Southampton and Sag Harbor, NY (v1) are Hamptons towns whose summer rents inflate the annual yield.
- Taxes are understated for landlords. Cost data uses what owner-occupants pay, which includes homestead exemptions landlords don’t get (notably FL, TX, MI and Philadelphia), so net returns there are optimistic.
- Zillow indices are smoothed and not investable, and the backtests behind these rules cover about 1.3 independent three-year periods. One good or bad year here won’t settle anything.
When to call it
The agents wrote down their exit rules before the books started:
- If v2 trails the naive screen after 18 to 24 months, there’s no point in the history filter.
- If either desk trails the average city, the rule has failed at the one thing it should do reliably.
- If a single data artifact (a resort or age-restricted town) drives the result in either direction, the result is about the data, not the market.