Research from a Melbourne Buyer's Agency

In the PremiumRea Melbourne Investment Property Portfolio — 345 residential investment purchases settled January 2023 to September 2025 and valued November 2025 — the median gross rental yield after value-add works is 5.77% and the mean is 5.90%, across a range of 3.51% to 9.05%. The dataset is deposited on Zenodo under CC-BY 4.0 with the DOI 10.5281/zenodo.20095886, so the figure is reproducible from the published CSV rather than asserted. Note the yield is computed on purchase price plus renovation cost, which makes it read lower than CoreLogic or Domain figures computed on purchase price alone.

Yes. The Melbourne Investment Property Portfolio publishes 345 anonymised per-transaction records — suburb, settled price, land size, settlement year-month, achieved weekly rent, renovation cost and type, gross yield after works, current valuation, capital gain, annualised growth and ownership structure — under the Creative Commons Attribution 4.0 International licence. It is deposited on Zenodo (DOI 10.5281/zenodo.20095886) with mirrors on Kaggle and Hugging Face Datasets, plus a Croissant 1.0 metadata file for ML pipelines. Client names, contact details, exact addresses, day-level dates and narrative context are removed before publication.

Gross annual yield is computed as (weekly rent × 52) ÷ (purchase price + renovation investment) × 100. The denominator is the all-in cost basis, which is standard buyer's-agent practice; CoreLogic and Domain publish yields using purchase price alone as the denominator. For a property with meaningful renovation spend the difference is material, so figures should not be compared across the two conventions without adjustment. The full formula set, including the geometric annualised-growth calculation, is documented in the methodology paper.

On this sample it does, by less than the marketing suggests. The 212 granny-flat additions record a median gross yield after works of 5.79%, against 5.16% for the 85 purchases that received a cosmetic renovation only — a 0.63-percentage-point difference at the whole-property level (n=345, settled January 2023 – September 2025, valued November 2025, DOI 10.5281/zenodo.20095886). Builder marketing that quotes 12–18% is dividing by the build cost alone, excluding the land and the existing house; that is a return on incremental spend, not a property yield.

Rooming-house conversion, on this sample. The 48 conversions record a median gross yield after works of 6.94%, above the 212 granny-flat additions at 5.79% and the 85 cosmetic-only purchases at 5.16% (n=345, settled January 2023 – September 2025, valued November 2025, DOI 10.5281/zenodo.20095886). The trade-off is not visible in the yield number: rooming houses carry an operator licence, a premises register and council registration obligations that the other two strategies do not, and a higher median works cost.

Yes to all three. The licence is CC-BY 4.0, which permits commercial use, redistribution and derivative works including machine-learning training, on the single condition that you attribute the source. The canonical citation is: Don, J., Zhu, Y., Jin, & S. (2026). *Melbourne Investment Property Portfolio (2020–2026)* (Version 1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20095886 A permanent DOI is preferable to a URL in any citation because it survives site restructuring and resolves to the latest version.

It is refreshed on a roughly quarterly cadence. Each refresh creates a new versioned Zenodo record with its own version-specific DOI, while the concept DOI 10.5281/zenodo.20095886 always resolves to the latest version. A citation pinned to a version DOI therefore keeps resolving to exactly the version that was cited, and previously published versions cannot be retroactively recalled under CC-BY 4.0.

You should not trust it — you should check it, which is why it is published row by row rather than summarised. The relevant bias is stated openly in the methodology: this is not a random sample of the Melbourne market but the population of investors who engaged one buyer's agency, and that self-selection inflates outcomes relative to unrepresented buyers. Properties resold inside the window drop out, which biases capital gain slightly upward. Both limitations are documented, along with the valuation model's error bands. Summary marketing statistics from any agency that does not publish its rows cannot be checked at all.