Would you have made more money
following this model?
We re-run the model as of each year since 2016 — using only the data that existed at the time — then check what actually happened to rents over the following 3 years. No hindsight, no cherry-picking: every year, every market.
Go back to a past year, say 2016, and rebuild the rankings using only the data that existed then: employment, population, building permits, and observed rents across all 53 metros.
Take the model's top quartile, its 13 highest-ranked markets that year. Think of it as the buy list the model would have handed you. The bottom quartile is its avoid list.
Let 3 real years pass. Measure the actual rent growth every market delivered, using a national observed-rent index. Nothing is adjusted after the fact.
Did the buy list out-grow the avoid list? Each rewind year is one test year — 2016 through 2022, seven tests in all. The buy list won all seven.
Extra rent growth captured by the model's favorites, every test year
Each bar: markets the model ranked in its top quartile that year out-grew its bottom quartile by this much, per year, over the following three years.
Every test year, in full
| Ranked in | Measured over | Top quartile grew | Bottom quartile grew | Edge | Hit rate |
|---|---|---|---|---|---|
| 2016 | 2016–2019 | 6.2%/yr | 3.3%/yr | +2.9%/yr | 85% |
| 2017 | 2017–2020 | 5.3%/yr | 3.4%/yr | +1.9%/yr | 69% |
| 2018 | 2018–2021 | 8.9%/yr | 5.4%/yr | +3.5%/yr | 100% |
| 2019 | 2019–2022 | 10.7%/yr | 8.5%/yr | +2.2%/yr | 77% |
| 2020pandemic-era | 2020–2023 | 9.6%/yr | 7.8%/yr | +1.7%/yr | 62% |
| 2021pandemic-era | 2021–2024 | 5.8%/yr | 5.5%/yr | +0.3%/yr | 46% |
| 2022 | 2022–2025 | 2.6%/yr | 2.4%/yr | +0.2%/yr | 54% |
The model's five highest-conviction markets in each test year, with the rent growth each one actually delivered per year over the following 3 years. A check means the pick out-grew the median market that year.
Why do 2020–21 look weaker?Rankings produced in those years were built on pandemic-distorted trailing data — eviction moratoria, stimulus, and a once-in-a-generation migration shock. The model's edge narrowed but stayed positive, and it visibly recovers in the 2022 test year as inputs normalize. We'd rather show you that than hide it.
What +2.1% per year means:on a $50M portfolio growing NOI at market pace, selecting from the model's top quartile historically added roughly two extra points of rent growth per year — compounding to ~11% more cumulative growth over a five-year hold, before any operational upside.
For the quantitatively inclined: average rank correlation (Spearman ρ) between model score and realized 3-yr rent growth is 0.56 across normal-market test years. Category weights were fit on the 2016–18 tests and validated strictly out-of-sample on 2019–22; later input additions were pre-registered — never grid-searched — and accepted only after improving those same held-out tests. The model is frozen as v0.5.0 and each future quarter is a live, unretouched test. Rent outcomes measured by a national observed-rent index.