Property income may be underestimated (SPI ~£26bn vs HMRC ~£48bn)

[MaxGhenis]

Summary

Property income calibration targets from HMRC SPI (Survey of Personal Incomes) public tables may underestimate actual UK rental income by approximately 2x compared to HMRC's dedicated Property Rental Income Statistics.

Evidence

Source	Property Income	Year	Notes
HMRC SPI tables (current calibration target)	~£25.8bn	2020-21	Via incomes.csv from spi.py
HMRC Property Rental Income Statistics	£47.62bn	2023-24	Dedicated property income analysis

Sources:

• Current: HMRC Investment Income Statistics (SPI-based)
• Better source: HMRC Property Rental Income Statistics 2024

Why the gap?

Both sources are from HMRC, but:

• SPI tables: Aggregated from self-assessment tax returns - may miss smaller landlords below reporting thresholds
• Property Rental Statistics: More comprehensive administrative data specifically focused on property income

This is similar to how SPI savings interest (~£3bn) dramatically underestimated ONS D.41 household interest (~£98bn) - see PR #229.

Data lineage

The calibration flow is:

1. spi.py reads HMRC SPI publications → creates incomes.csv
2. incomes_projection.py uses incomes.csv + SPI_2020_21 dataset → creates incomes_projection.csv
3. loss.py reads incomes_projection.csv → targets labeled hmrc/property_income_*

Recommended action

1. Add HMRC Property Rental Income Statistics as calibration target (similar to ONS D.41 fix in PR Fix savings income calibration to match ONS D.41g #229)
2. Consider adding comment/docstring clarifying that incomes.csv is derived from HMRC SPI tables (taxable income only)
3. Review other income types for similar gaps between SPI and dedicated HMRC statistics

Context

This was identified while investigating savings income underestimation (PR #229), where SPI showed ~£3bn vs ONS ~£98bn. Property income underestimation is less severe (2x vs 33x) but still notable.

[nwoodruff-co]

Not sure about the HMRC stats here, could our income.csv be wrong? @PolicyEngine can you check the official HMRC unearned income stats publication

[policyengine]

I've verified the HMRC official statistics on property income. Here's what I found:

Official HMRC SPI Investment Income Statistics (2020-21)

I downloaded and parsed the official HMRC Table 3.7 for 2020-21 from the Investment Income Statistics publication.

Property income total: £24.5 billion (£24,500 million)

This matches what's in the current incomes.csv file (£25.8bn is close - there may be minor rounding or the file uses a slightly different year's data).

HMRC Property Rental Income Statistics

The Property Rental Income Statistics 2024 publication shows:

Property income for 2020-21: £46.68 billion

Why the difference?

Both are official HMRC statistics, but they measure different things:

1. SPI Investment Income Statistics (£24.5bn): Based on the Survey of Personal Incomes, covering "individuals with some liability to tax" (Note 7 in the publication). This is derived from tax return data and focuses on taxable income.

2. Property Rental Income Statistics (£46.68bn): More comprehensive administrative data covering all "unincorporated landlords declaring via Income Tax Self Assessment returns." This publication explicitly states it includes landlords declaring property income, whether or not they have tax liability.

Key differences explaining the gap:

• The SPI tables only include taxpayers with "some liability to tax"
• The Property Rental Income Statistics include all ITSA declarations, even those with no final tax liability (e.g., due to allowances, reliefs, or losses)
• Many landlords have rental income but no net tax liability after allowances

Verdict

The incomes.csv file is accurate for what it claims to be: income statistics from the Survey of Personal Incomes covering individuals with tax liability. However, for calibration purposes, you may want to consider using the Property Rental Income Statistics (£46.68bn for 2020-21) as it's more comprehensive and specifically designed for property income analysis.

This is analogous to the savings interest issue in PR #229, though the gap is smaller (1.9x vs 33x).