South African section 20A ring-fencing of assessed losses for natural persons, and the section 20 carry-forward rules it sits on. Covers the income threshold that triggers ring-fencing and the change from the 45% to the 39% marginal rate from 1 March 2026, the list of suspect trades, the three-out-of-five-years test, the escapes in section 20A(3) and the letting exception, and the company 80% limitation under section 20. Use for questions about ring-fenced losses, suspect trades, hobby farmin...
Installs into .claude/skills of the current project.
Are you the author of Za Assessed Loss Ring Fencing?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/openaccountants-za-assessed-loss-ring-fencing)
---
name: za-assessed-loss-ring-fencing
description: "South African section 20A ring-fencing of assessed losses for natural persons, and the section 20 carry-forward rules it sits on. Covers the income threshold that triggers ring-fencing and the change from the 45% to the 39% marginal rate from 1 March 2026, the list of suspect trades, the three-out-of-five-years test, the escapes in section 20A(3) and the letting exception, and the company 80% limitation under section 20. Use for questions about ring-fenced losses, suspect trades, hobby farming, rental losses, side businesses, or offsetting a trade loss against salary."
license: "AGPL-3.0-or-later (code) / OpenAccountants Guide License v1.0 (content)"
metadata:
source: openaccountants
openaccountants_url: "https://www.openaccountants.com/skills/za-assessed-loss-ring-fencing"
jurisdiction: "ZA"
last_updated: "2026-09-03"
reviewed_by: "Brandon Iverach"
---
# South Africa — Ring-fencing of Assessed Losses, s 20A
> **What changed, and why this file exists.** For years of assessment commencing on or
> after **1 March 2026**, s 20A bites from the **39%** marginal rate, not the 45% maximum
> marginal rate. The threshold roughly halves. Ring-fencing stops being a
> high-income edge case and becomes the **default** for a geared residential rental held
> by a reasonably paid employee.
---
## Section 1 — Scope
This file covers **s 20A ring-fencing for natural persons**, and enough of **s 20** to
place it.
Not covered: assessed losses of trusts; the s 20(1)(a) company rules beyond the summary in
Section 6; capital losses, which are an Eighth Schedule matter and not assessed losses at
all.
---
## Section 2 — The two-stage test
Ring-fencing under s 20A applies only where **both** stages are met.
**Stage 1 — the income test.** The taxpayer's taxable income, *before* setting off the
loss in question, equals or exceeds the amount at which the relevant marginal rate begins.
**Stage 2 — the trade test.** Either
- the trade is a **listed suspect trade** (Section 4), **or**
- the trade has made an assessed loss in **at least three of the last five** years of
assessment, counting the current year.
If stage 1 fails, s 20A does not apply at all, whatever the trade. If stage 1 is met, a
listed suspect trade is caught **immediately** — it never needs the three-in-five test.
_Source: Income Tax Act 58 of 1962 s 20A(2)._
---
## Section 3 — The income threshold, by year of assessment
The threshold is **not a fixed rand amount**. It is defined by reference to a marginal
rate, so it moves every year with the bracket table.
| Year of assessment | Rate the threshold keys off | Taxable income threshold |
| --- | --- | --- |
| 2023 (2022/23) | 45% — maximum marginal rate | R1,731,600 |
| 2024 (2023/24) | 45% — maximum marginal rate | R1,817,000 |
| 2025 (2024/25) | 45% — maximum marginal rate | R1,817,000 |
| 2026 (2025/26) | 45% — maximum marginal rate | R1,817,000 |
| **2027 (2026/27)** | **39%** | **R695,800** |
**Do not hard-code the rand figure.** Read it off the bracket table for the year in
question as the floor of the band carrying the relevant rate. See `za-income-tax-tables`.
⚠️ **A common error in commentary written during 2025** quotes the new threshold as
**R673,000**. That was the 39% bracket floor for the *2026* year of assessment, current
when the amendment was proposed. Because the amendment only takes effect for years of
assessment commencing on or after 1 March 2026, the first year it applies is 2027, whose
39% floor is **R695,800**. Using R673,000 for 2027 is wrong.
_Source: Income Tax Act 58 of 1962 s 20A(2)(a), as amended with effect for years of
assessment commencing on or after 1 March 2026; SARS Rates of Tax for Individuals for the
bracket floors._
---
## Section 4 — The listed suspect trades
Under s 20A(2)(b), these are caught on the income test alone:
- Any sport practised by the taxpayer or a relative
- Any dealing in collectibles
- **The rental of residential accommodation**, unless the letting exception in Section 5
applies
- The rental of vehicles, aircraft or boats
- Animal showing by the taxpayer or a relative
- Farming or animal breeding, unless carried on on a full-time basis
- Any form of performing or creative arts
- Any form of gambling or betting
- Any trade in respect of which a tax benefit scheme applies
_Source: Income Tax Act 58 of 1962 s 20A(2)(b)._
---
## Section 5 — The escapes
A trade that is otherwise caught is **not** ring-fenced where one of these applies.
**The letting exception — s 20A(2)(b)(iii).** Residential accommodation is not a suspect
trade where **at least 80%** of the accommodation is used by persons who are **not
relatives** of the taxpayer for **at least half** the year of assessment. Note both limbs:
the 80% is about who occupies it, and the half-year is about duration.
**Reasonable prospect of profit — s 20A(3).** The trade escapes if it constitutes a
business with a reasonable prospect of deriving taxable income within a reasonable period,
having regard to the s 20A(3) factors — among them the proportion of gross income to
allowable deductions, the commercial manner in which the trade is carried on, the number
of years of losses, the business plan, and whether assets are available for private use.
This is a facts-and-evidence test, not an assertion. A business plan, arm's length
pricing, separate banking and a credible route to profitability are what carry it.
_Source: Income Tax Act 58 of 1962 s 20A(2)(b)(iii), s 20A(3)._
---
## Section 6 — Section 20, for context
Ring-fencing modifies an ordinary carry-forward. The underlying rules:
- An assessed loss carries forward and may be set off against income from **any** trade,
subject to s 20(2A), which requires the taxpayer to have carried on a trade in the year
of set-off.
- **Companies** are separately limited: for years of assessment commencing on or after
1 April 2022, the balance of assessed loss that may be set off is capped at the **greater
of R1,000,000 and 80%** of taxable income before the set-off. **This 80% cap does not
apply to natural persons.**
_Source: Income Tax Act 58 of 1962 s 20, s 20(2A)._
---
## Section 7 — What ring-fencing actually does
A ring-fenced loss is **not disallowed**. It is quarantined: it may only be set off against
income from **that same trade** in future years, not against salary, not against other
trades.
So the cash effect is a deferral, sometimes indefinite. A rental that never turns a taxable
profit carries a ring-fenced loss forward permanently without ever relieving anything.
---
## Section 8 — Edge cases
- **Test the income threshold before anything else.** Below it, s 20A is irrelevant and no
amount of suspect-trade analysis matters.
- **"Taxable income before the set-off"** is the measure — not gross income, not
remuneration, and not taxable income after the loss.
- **A geared residential rental is now the standard case, not an exotic one.** A salaried
taxpayer over the 39% threshold with a bonded rental producing an interest-driven loss is
ring-fenced by default from 2026/27 unless the letting exception or s 20A(3) applies.
Practitioners who learned this section when it began at 45% will misjudge it.
- **Relatives occupying the property** count against the 80% limb of the letting exception.
Letting to a child at market rent still fails it.
- **Farming is caught only where it is not full-time.** Full-time farming falls outside the
list, and has its own regime in the First Schedule.
- **The election is not annual.** Once ring-fenced, the loss stays ring-fenced to that trade
until it is absorbed by profits of that trade.
---
## Section 9 — Self-checks
1. **Threshold from the right year's bracket table.** Confirm the taxable income threshold
is the floor of the band carrying the relevant rate for that year of assessment — not a
figure carried over from a prior year, and not R673,000 for YoA 2027.
2. **Right rate for the right year.** 45% for years of assessment up to 2026; 39% from
2027. Applying 39% to a 2026 return, or 45% to a 2027 return, both give the wrong answer.
3. **Suspect trade first, three-in-five second.** If the trade is listed, the three-in-five
test is irrelevant. Running it anyway produces a false negative.
4. **Both limbs of the letting exception.** 80% non-relative occupation **and** at least
half the year. One without the other fails.
5. **Ring-fenced, not lost.** Confirm the loss is carried forward against that trade and
not written off.
---
## Disclaimer
> **General reference only.** This file is general tax reference material for AI-assisted
> workflows. It has not been reviewed for any specific person's facts, documents,
> elections, deadlines, residency, filing status or local procedures. Do not rely on it to
> file, pay, amend or take a tax position without review by a qualified professional in
> South Africa.
> Contributed by Brandon Iverach.
## Today's figures and who stands behind this skill
This file is an AI skill (tax rules your AI follows) from OpenAccountants, copied on 2026-09-03. Rates, thresholds and deadlines change. For the current version of this skill, its official sources and the named accountant (if any) who stands behind it, call `get_skill` with slug `za-assessed-loss-ring-fencing` on the OpenAccountants connector (https://www.openaccountants.com/connect), or read https://www.openaccountants.com/skills/za-assessed-loss-ring-fencing. If neither is available, confirm every figure against the official source pages cited above before relying on it.