TL;DR: Most apartment rent and occupancy forecasts are built top-down—they start with a national or metro-level estimate and divide it across smaller geographies. ApartmentIQ does the opposite. We forecast rent growth and occupancy for individual properties first, then aggregate those property-level forecasts—weighted by unit count—up to the submarket, ZIP code, and metro level. The result is a forecast that reflects how real assets actually behave, preserves the differences between a Class A high-rise and a Class C garden community in the same market, and updates from publicly available listing data in near real time.
What is bottom-up rent forecasting?
Bottom-up rent forecasting is a method that generates a rent and occupancy forecast for each individual apartment property first, then builds market-level forecasts by aggregating those property forecasts together. The metro number isn’t an assumption handed down from the top—it’s the sum of the real assets inside that metro, weighted by how many units each one contains.
This is the reverse of the conventional approach. A top-down model begins with a broad market estimate—often national or metro—and disaggregates it downward to smaller areas using ratios or historical shares. Bottom-up modeling starts at the most granular level that actually transacts rent: the individual community.
How is bottom-up different from top-down forecasting?
The distinction matters because apartment performance is intensely local. Within a single metro, a newly delivered luxury lease-up and a stabilized 1980s value-add asset can move in opposite directions in the same quarter. A top-down model smooths over that heterogeneity; a bottom-up model is built from it.
| Top-down forecasting | Bottom-up forecasting (ApartmentIQ) | |
|---|---|---|
| Starting point | National or metro aggregate | Individual property |
| Direction | Disaggregates down to submarkets | Aggregates up to metro |
| Local detail | Smoothed into market averages | Preserved per asset, class, and unit type |
| Granularity | Typically metro or submarket | Property → ZIP → submarket → metro |
| Best for | Broad macro context | Asset selection, underwriting, portfolio monitoring |
Because every higher-level forecast is an aggregation of the assets beneath it, the same engine can answer “How will this specific community perform?” and “How will this entire metro perform?”—and the two answers stay internally consistent.
How does ApartmentIQ forecast rent and occupancy?
The model predicts two variables that are the most direct, actionable indicators of apartment market health:
- Rent growth — year-over-year percent change in asking rent
- Occupancy — share of units occupied
Both are derived from publicly available apartment listings, collected and maintained by ApartmentIQ. That sourcing is a deliberate advantage. Listing data updates continuously and reflects current market conditions in near real time. Unlike surveys or closed-transaction data—which can lag by months—listings capture asking rents and availability as they’re actually presented to prospective renters, providing a leading view of where rents and occupancy are heading.
At the core is a deep learning model purpose-built for multi-horizon time series forecasting. It combines recurrent layers (LSTM) to capture short-range momentum and seasonal patterns with attention mechanisms that identify longer-range cycles—multiyear rent cycles, permit-to-delivery lags, and similar structural relationships. Gated residual networks and variable selection networks let the model focus on the most predictive signals and suppress noise, which improves generalization. Rent growth and occupancy are predicted jointly within a single model, so it can exploit the correlation between the two rather than treating them as independent.
What data feeds the forecast?
The model layers local listing signals on top of a broad set of economic covariates, organized by geographic tier:
- National: nonfarm payroll employment, inflation (CPI), retail sales, and population (sources: U.S. Bureau of Labor Statistics and U.S. Census Bureau)
- Metro: average weekly earnings, unemployment rate, and building permits (new residential units authorized)
- Submarket: new construction deliveries—the actual supply of new units entering a local market
The supply signals are especially important. Building permits foreshadow competition for renters one to two years out, while new deliveries measure supply that’s hitting the market right now. Markets absorbing large volumes of new supply tend to see softer occupancy and slower rent growth; markets with limited new supply tend toward tighter occupancy and stronger pricing power.
How accurate are the forecasts?
Accuracy is validated out-of-sample: the model is trained on history, a recent period is withheld, and the forecast is graded against what actually happened. Against 2025 results, ApartmentIQ’s rent growth forecasts were roughly 0.7 percentage points more accurate than the publicly available forecasts the industry relies on—a 33% improvement in mean absolute error. With total annual market movement typically just 3–4%, that margin is a meaningful edge for pricing and underwriting decisions.
We break down the full benchmark, including a market-by-market comparison across major MSAs, here: How ApartmentIQ Builds Forecasts You Can Trust.
Why provide three scenarios instead of one number?
No forecast is certain, and a single point estimate can create a false sense of precision when markets are subject to economic cycles, policy changes, and local supply shocks. For every geography and every month across the five-year horizon, the model produces three scenarios:
- Baseline — the central estimate (the median, or 50th percentile), reflecting the most likely outcome
- Upside — a more favorable outcome with stronger demand or tighter supply (75th percentile)
- Downside — a weaker outcome with softer demand or elevated supply (25th percentile)
These come directly from the model’s probabilistic (quantile) outputs, so the spread between upside and downside reflects genuine uncertainty—wider in volatile markets and longer time horizons, tighter where conditions are stable. That lets investors stress-test underwriting assumptions and evaluate downside exposure rather than relying on a single line.
What can you actually do with property-up forecasts?
Because the forecasts run from the property all the way to the metro—monthly, over a five-year horizon, in three scenarios—they support a range of workflows:
- Acquisition underwriting: translate forecasted rent growth and occupancy into projected revenue, NOI, and valuation at the asset level.
- Market selection and capital allocation: compare expected performance across metros and submarkets on a consistent, data-driven basis.
- Portfolio monitoring: track how the outlook for owned assets evolves and flag properties or markets where conditions have materially shifted.
- Risk management: use the upside/downside range to size exposure and pressure-test the base case.
Frequently asked questions
What is the difference between top-down and bottom-up rent forecasting?
Top-down forecasting starts with a national or metro estimate and divides it across smaller areas. Bottom-up forecasting starts with individual properties and aggregates them up. ApartmentIQ uses bottom-up modeling, so each metro forecast is the unit-count-weighted sum of the real properties inside it.
What does ApartmentIQ forecast?
Year-over-year rent growth and occupancy, delivered monthly over a five-year horizon, in baseline, upside, and downside scenarios, at the property, ZIP code, submarket, and metro levels.
Where does the underlying data come from?
Rent and occupancy are derived from publicly available apartment listings updated continuously, combined with macroeconomic indicators from the U.S. Bureau of Labor Statistics, the U.S. Census Bureau, and third-party construction data.
How accurate is the model?
Against 2025 results, ApartmentIQ’s rent growth forecasts were about 0.7 percentage points more accurate than publicly available forecasts—a 33% improvement in mean absolute error, validated out-of-sample. See the full benchmark in How ApartmentIQ Builds Forecasts You Can Trust.
Why does forecasting at the property level matter?
Apartment performance varies sharply by asset class, unit type, and location—even within one metro. Forecasting each property first preserves those differences, which is critical for asset selection and underwriting rather than just broad market context.