Why it feeds property value

Each HQ relocation brings a slug of employment income into the metro, which flows into the affordability anchor and then into the price line. The Goldman Sachs campus alone represents 5,000+ potential households entering a single ZIP’s demand pool. The effect is concentrated: Uptown and the North Dallas office corridor absorb most of the headcount, which is why the 75201–75204 band has appreciated faster than the county.

The shift from full relocations to regional hubs is worth noting. A hub brings the jobs but not the C-suite — which means local office demand without the kind of marquee announcement that bids up land ahead of it. That is, arguably, the better trade for an investor: the demand arrives without the pricing signal.

Where the momentum is slowing

CBRE’s own report notes that DFW’s lead is narrowing. Miami, Charlotte and Nashville are gaining share, and rising Dallas home prices — the same ones that fill this portal — are among the factors site selectors now cite against automatic DFW recommendations. Competing metros offer newer office stock and, in some cases, lower effective housing costs.

The other risk is concentration in a single story. A large share of the demand traces to employers leaving California and New York for tax and regulatory reasons. That flow is policy-sensitive: a change in either the origin states or in Texas’s own cost structure could slow or reverse it in ways that organic population growth would not.

Why it matters for property

Roughly $4.6B of committed public capital — convention centre, Fair Park, the 2024 bond — is landing on a county where the median ZIP still trades around a quarter of a comparable Los Angeles address. Most of it is concentrated in a narrow band: the southwest corner of downtown, Fair Park and South Dallas, and the East Dallas drainage corridor. Those are, not coincidentally, among the lowest-priced ZIPs on the map.

The funding structure is the part worth internalising. Hotel tax and general obligation bonds do not raise your rate; the appraisal does that on its own. So the public programme is, for an investor, close to a free option — upside in the value line with no matching entry in the cost line. That is the opposite of how civic investment usually works, and it is a direct consequence of Texas funding its cities the way it does.

The honest risk

The convention centre is not fully financed. As of early 2026 the city was extending a $1B bridge loan and had pushed its revenue bond issuance to the autumn, with $665M of the plan resting on naming rights and asset sales that have not been realised. Hotel tax receipts are cyclical, and the project finance zone assumes visitor growth that a downturn would not deliver.

Underneath that sits a subtler problem. Dallas County is not the only place capturing DFW’s growth — Collin and Denton have been taking a rising share of it, with newer stock and, in places, lower effective rates. A civic programme concentrated downtown is a bet that the county centre holds its pull. Watch the population-change layer on the ZIP map: several inner-county ZIPs have been shrinking while the metro adds a million people a decade.

Investor-favourable

No state income tax. Rental income and capital gains face federal tax only. For a high-bracket owner this is worth more than most state-level real estate incentives, and unlike Oklahoma’s holding-period capital gains deduction it carries no conditions, no holding period and no sunset risk.

No rent control, and no path to one. State law preempts municipal rent control. Rents reset at market on every renewal, and there is no local political route around that.

Fast, cheap eviction relative to most large metros. Texas justice courts move quickly and filing costs are low. This is a real component of a defensible vacancy assumption, not a talking point.

Annual appraisal protest. Because appraisals reset yearly they can be contested yearly, with a mid-May deadline in Dallas County. A successful protest is a permanent reduction to the largest line in the P&L, and it compounds over a hold.

Elastic supply keeps entry sane. Dallas builds. That caps appreciation, but it also means you are rarely paying a scarcity premium, and construction cost rather than land sets the floor.

Watch closely

The homestead sequence, and what it excludes. Five exemption increases in a decade — $15k in 2015, $25k to $40k in 2021, $40k to $100k in 2023, $100k to $140k in 2025. Every one of them is a homestead measure. An investment property receives none of them, which means the relief has been narrowing the base while your share of it grows.

The one measure that did reach investors expires. The 2023 package included a three-year pilot capping appraisal growth at 20% a year on non-homestead property under $5M. It is a pilot. Check its status before you model a hold that depends on it.

Insurance. North Texas hail has repriced property insurance sharply. Roof age drives quotes, and a renovation that skips the roof can leave a policy uninsurable or priced punitively.

Short-term rental rules are unsettled. The city’s 2023 ordinance restricting short-term rentals in single-family districts has been in litigation. Do not underwrite a Dallas deal on short-term rental income without current legal advice.

Structural risk

The appraisal ratchet. Your rental is reappraised to full market value every year with no cap. In a rising market the tax line grows at the appreciation rate, not at inflation — a five-year hold at 5% annual appreciation raises the bill by roughly 28%. Model it as a percentage of value, never as a fixed dollar amount, which is why this portal charges it against the modelled value each year.

School finance is outside local control. The ISD share is the largest component of the bill and is set by state formula, recapture and compression. There is no one to lobby locally.

Non-disclosure state. Texas does not require sale prices to be public. Every price series here, Zillow’s included, is modelled from listings and tax records rather than observed from a register. Confidence intervals are genuinely wider than in a disclosure state.

Concentration in corporate relocation. A large share of recent demand traces to employers moving headcount out of higher-cost states, a flow that is policy-sensitive and reversible in a way organic population growth is not.

Why error correction

A pure momentum model extrapolates a boom forever; a pure affordability model calls every hot market a bubble the year before it doubles. An error-correction specification carries both: short-run dynamics from momentum and flow variables, plus a long-run anchor that pulls price back toward what local income can actually finance.

The strength of that pull is λ, and it is estimated below rather than assumed. Dallas comes out with a small but statistically real λ — the market has repeatedly run ahead of local income and then cooled rather than corrected.

Where P* differs from the OKC build

In a low-property-tax market you can approximate the affordability anchor with principal and interest alone. In Dallas you cannot: tax and insurance are roughly a third of the monthly payment on a median home. Both are therefore inside P*.

The consequence is worth naming. Because a large slice of the Dallas payment does not respond to interest rates, a given rate cut buys less affordability here than the same cut buys in a low-tax metro. The rate slider will show you this directly.

The IRR simulator

Monthly cash flows, solved for the internal rate of return by Newton–Raphson and annualised. Equity in is down payment plus renovation plus every buy-side cost. Operating flows carry rent growth, vacancy, operating costs, property tax on the modelled value, insurance, and debt service split into interest and principal. Exit nets commission, closing, title and the loan payoff.

Property tax is charged against the model's projected value each year, not the purchase price. That is how Texas actually works, and it is the difference between a plausible five-year IRR and a flattering one.

The advisor

Each ZIP is ranked by percentile on every criterion you select, with direction handled per criterion — high yield is good, high price per entry is not. The composite was originally the simple average of those percentiles, and by default it no longer is.

The simple average had a real failure mode. A percentile rank has no way to distinguish a genuinely tight market from a market too small to generate many listings in the first place. Wilmer (75172, population 5,800) topped an early version of this ranking almost entirely because it has 12 active listings against a county median of 111 — “tight supply” was measuring market size, not market tightness. That is a statistical artifact, not an investment signal, and a page that presents a #1 ranking has an obligation not to hand you one.

Three fixes, all visible. First, the supply criterion is now listings per 10,000 residents rather than a raw count, which removes the size bias directly. Second, every ZIP carries a confidence weight built from its population percentile, its listing-count percentile, and whether its rent data trips the divergence flag described in the tax drag section. In confidence-weighted mode — the default — each ZIP’s composite score is pulled toward the county-neutral midpoint by an amount proportional to (1 − confidence²), so a thin market needs a decisively stronger raw signal to reach the top than a deep one does. The squared term matters: linear shrinkage let a moderate-confidence ZIP like Mesquite (75181, pop 30k) coast into a top-five position on a decent raw score. Squaring pushes the top of the list to lean high-confidence — typically nine of twelve top ranks under any four-criterion mix. Wilmer moved from #1 to #19 of 75 when this was applied; the county’s largest ZIP by population, 75217, took its place with a real 10.5% gross yield on 85,000 residents and 132 listings.

Third, exclusion rather than fabrication. Any ZIP missing data on a selected criterion is removed from the ranking outright and listed in a separate table below the map, showing exactly which criterion it lacks. This applies in both scoring modes: the 15th-percentile floor on missing data is only for the confidence weight, never for the underlying signal. If Zillow doesn’t publish a ZHVI for a ZIP, that ZIP cannot appear in any yield ranking. Change the criterion mix and the excluded set changes with it.

The raw-average mode is kept, not hidden, behind a toggle above the map — because shrinkage is itself a judgment call, and you may have reasons to want the unweighted picture (comparing a known small market against itself over time, for instance, where the size bias is constant and therefore harmless). The confidence badge in every advisor popup and in the ranking table tells you which regime you are looking at either way.

What the fit refused to give

The interest rate has no separately identified short-run term. Every specification tried — contemporaneous, lagged one to three months, cumulative over six and twelve months — either came out statistically indistinguishable from zero or, worse, positively signed. That is not evidence that rates do not matter. It is what happens when rates rise in exactly the periods when the economy is hot and prices are climbing anyway: 2004–06 and 2021–22 both sit inside the sample.

So the rate was left to act the way theory says it should — by moving the affordability anchor, which then pulls on price through λ. The decomposition chart separates that rate-driven part of the pull from the rest, so you can still see the channel working. Calibrate your trust in it with this: on the county median home, a two-point rate rise takes about 6% off the modelled twenty-four-month price and turns annualised growth from roughly +1.7% to −0.4%. That is a real elasticity, but it is slower than the 2022 experience, when rates moved four points and price growth stalled within months. The model books most of that stall as momentum decay rather than as a rate shock, which is a limitation you should hold in mind whenever you move the slider hard.

What could not be estimated at all

Population and income growth. Regressed annually against price growth over twenty-four years, both come back insignificant with an R² of 0.01. The reason is not that they do not matter — it is that DFW population growth has sat near two percent every single year, so there is almost no variation to identify against. Their long-run contribution is instead absorbed by the fitted intercept.

Permit flow. Correctly signed but nowhere near significance. Retained at its fitted value so the supply slider does something, but it does very little, and you should not read much into it.

The population and income sliders therefore move the forecast as deviations from the fitted baselines — 2.0% and 2.5% a year. Leave them at baseline and the model runs on estimated parameters alone. Move them and you are layering a judgement on top, which is a legitimate thing to do as long as you know that is what you are doing.