Multifamily - Interactive Market Prediction Tool

Austin built more apartments in 2023 and 2024 than in any two years in its history, increasing market-wide vacancy from 6 to 16 percent. Use this tool to predict Austin's path back to equilibrium, using the variables below* fitted on 25 years of Austin's CoStar data and checked against the 37 largest U.S. metros. Use the toggles and movable line graphs below to the left to experiment with different scenarios.

    ▶ Methodology: click to expand

    The tool is three formulas run forward one year at a time, from the latest CoStar quarter to 2036. Each year, in order: how much gets started, how much gets leased, and what that does to rents. The next year begins from where this one ended. Every number in the formulas comes from the data; nothing is hand-set.

    What feeds what

    FormulaDrivers (and their timing)What it captures
    Starts
    % of stock
    Last year's rent growth · last year's 5-year Treasury · last year's starts · last year's absorptionDevelopers build after rents rose, cut back when money got dearer, keep about half of last year's pace (pipelines and lenders do not turn on a dime), and build where leasing was strong. Everything is one year back, because a start decided this year was underwritten on last year's numbers.
    Deliveries
    units
    Units under construction today (the next two years) · then the average of starts one and two years earlierThe pipeline is known, so the next two years are close to certain; after that a building takes one to two years from start to delivery.
    Absorption
    % of stock
    This year's job growth · last year's rent growth · this year's deliveries · last year's absorption · the population settingJobs bring renters now. A big rent increase last year pushes people to double up or buy, with the year's delay that leases impose. New buildings lease up in their first year, but a flood leases more slowly than a trickle (the curve in the formula). Leasing has momentum. Population growth above or below the market's norm adds or subtracts renters over time.
    Vacancy
    %
    Last year's vacant units + deliveries − absorption, divided by the stockArithmetic, not a fit. A floor stops absorption of units that do not exist.
    Rents
    % growth
    This year's vacancy · the change in vacancy during the year · this year's income growth · inflationHigh vacancy makes landlords cut, low vacancy lets them raise; a market whose vacancy is rising cuts faster than one sitting still; income growth lifts what renters can pay; inflation is added back on top of the real relationship.

    Two fits for every slope, blended

    • The baseline: each formula fitted once across all 37 largest U.S. metros at the same time, 2002 to 2025, with each market measured against its own averages so that a 3% market and a 10% market contribute only their ups and downs. About 890 market-years, so the slopes are tight.
    • The local fit: the same formula fitted on this market's own 24 years alone. It carries the market's character but is noisy, and with a narrow history it can come out with the wrong sign.
    • The blend: slope used = baseline + w × (local − baseline). The weight w runs from 0 to 1 and is computed, not chosen: it is high when the local slope is precisely estimated and markets genuinely differ on that driver, low when the local history cannot pin it down. Each slope gets its own w. today: .
    • The constants are always local: once the slopes are set, each market's constant is solved so the formula passes through its own averages. That keeps the balance-point vacancy, the normal pace of building and the typical year's absorption the market's own; the baseline supplies only the shape of the response.
    • One rule: if the blend still contradicts the baseline's sign, the slope stays at the baseline and the page says so under the formulas.

    Lags

    Most drivers enter with a one-year lag (last year's rent growth, last year's Treasury, last year's starts and absorption), because the decisions they drive are made on last year's information and leases and construction take time to show up. Job growth, deliveries, vacancy and income growth enter in the same year, because they act on renters and landlords as they happen. The lags were not assumed: one, two and three-year versions of each driver were tested, and the one-year versions predicted best on years the formulas had not seen.

    Artificial and human intelligence

    This tool was built through a collaborative process between artificial and human intelligence. We use an iterative process of trial and testing, with AI doing the bulk of the heavy statistical work and people testing the results for logical interactions and results, suggesting changes and reiterating the process.

    ▶ * Variables and formulas: click to expand

    Source: CoStar Group, multifamily DataExport for 394 U.S. metros, Q1 2000 to Q2 2026, As Of 2026 Q2, annualized; Federal Reserve H.15 via FRED (GS5) for the 5-year Treasury.

    ▶ Back test: run the projection from an earlier year and compare it with what happened

    The formulas run forward from that year's actual position with the job growth, Treasury yields, inflation and income growth that actually happened, and the pipeline of the following two years as it was; starts, absorption, vacancy and rents are projected. Solid lines are actual, dashed are projected.

    Vacancy, actualVacancy, projectedEffective rent per SF, actualRent, projected

    Stabilized (under 8%)¹
    ·
    Reaches your target¹
    ·
    Vacancy now
    Excess vacant units
    Rent back to prior peak
    ·
      Scenario

      Vacancy rate and effective rent, history and projection

      Annual, year end, MSA. Vacancy in navy on the left axis; effective rent per square foot in red on the right axis. Solid = CoStar history, dashed = projection under your assumptions; the green line is your target and the gray line the frictional floor.

      Vacancy rate (left)Effective rent, $/SF per month (right)TargetFloorQ2 2026 actual

      Vacant units vs. target (lines) and net annual change (bars)

      Units, thousands. Lighter bars and the dashed line are projected.

      Vacant units at year endVacant units at your targetAnnual increase / decrease in vacant units (see chart below)Q2 2026 actual
      Deliveries (new units completed)Net absorption (units newly occupied)

      Year by year

      Projection rows in italics. Actual first-half 2026: .

      Sources

      • CoStar Group, multifamily DataExport for 394 U.S. metros, Q1 2000 to Q2 2026, As Of 2026 Q2, annualized: inventory, vacancy, effective rent, deliveries, net absorption and starts, one MSA per market in the dropdown. costar.com
      • Federal Reserve Board H.15 via Federal Reserve Bank of St. Louis (FRED), GS5 (5-year Treasury constant maturity yield), annual averages 2000 to 2025 and January to August 2026. fred.stlouisfed.org/series/GS5
      • 5-year Treasury forward curve, monthly resets September 2026 to September 2036, as of September 14, 2026, averaged by calendar year for the default rate path.
      • Federal Reserve Bank of St. Louis (FRED), T5YIE (5-year breakeven inflation rate) and T5YIFR (5-year, 5-year forward inflation expectation rate), September 11, 2026, for the default inflation path. fred.stlouisfed.org/series/T5YIE
      • Federal Reserve Bank of St. Louis (FRED), CPIAUCSL, December values, for the inflation-adjusted rent fit. fred.stlouisfed.org/series/CPIAUCSL