The performance test and the herding problem

The number that killed the deal was 0.3 percentage points. A superannuation fund's investment committee had spent two meetings on a proposal to commit roughly $300 million to a portfolio of Australian transmission and firming assets — poles, wires and grid-scale batteries, the physical equipment an electricity system running on wind and solar cannot work without. The projects were real. The counterparties were credible. The internal modelling showed a return well ahead of what the fund expected from listed shares over the same twenty-year horizon. Then the risk officer put up the last slide, and it was not about returns at all.

The slide showed what the allocation would do to the fund's score in the annual superannuation performance test. Not if the assets failed — if they performed exactly as modelled, while the listed infrastructure index they are measured against happened to run ahead of them for a few years. On that scenario the allocation could drag the product roughly 0.2 to 0.3 percentage points below where it would otherwise sit across the test window. The threshold for failing is half a percentage point. Nobody in the room thought the investment was bad. Several said, on the record, that they thought it was good. The committee deferred it, and then quietly let it lapse.

That committee is a composite. It is not a real one, and no real fund, executive or project is described in it. The arithmetic it faces is entirely real, it is written into Commonwealth regulations, and the rest of this article sets it out in full.

Two thousand kilometres away, a 24-year-old hospitality worker has never opened a superannuation statement in her life. She has changed employers four times. She could not name her fund without checking her phone, could not tell you what it charges her, and has never made a single decision about where her money is invested. She is not unusual. She is the system working as designed — compulsory, defaulted, and built on the assumption that most people will never engage. And she is the person the performance test exists to protect. Because of it, she is measurably less likely to spend her working life in a product that quietly underperforms. She will never know that happened.

Both of those people are consequences of the same policy. That is the whole problem, and it is a harder one than either side of the argument usually admits.

What the test is, and who legislated it

The annual superannuation performance test was created by the Treasury Laws Amendment (Your Future, Your Super) Act 2021, which received Royal Assent in June 2021 and took effect from 1 July 2021. It was legislated by the Morrison government. This matters and it should be stated at the top: the current government did not create this test. It inherited it. What it owns is everything since May 2022 — the decision to keep the test, the way it has been administered, and the decision to review its design.

The Act, known universally as YFYS, did four things at once, and they are designed to interlock. It "stapled" each worker to a single superannuation account that follows them from job to job, ending the old practice of collecting a new account with every new employer. It created a public comparison tool, YourSuper, hosted by the Australian Taxation Office. It strengthened the legal duty on trustees to act in members' best financial interests, with the burden of proof on the trustee. And it required the Australian Prudential Regulation Authority to run an annual performance test on superannuation products, with automatic consequences for failure.

The test itself works like this. APRA takes the product's actual net investment return — returns after investment fees and tax — over a rolling window, originally eight years and now ten. It then constructs a benchmark return for that same product. The benchmark is not a single market index. It is built from the product's own strategic asset allocation: whatever weights the trustee has chosen across Australian shares, international shares, fixed income, property, infrastructure, cash and so on, applied to a set of indices prescribed in regulations for each of those asset classes. The result is then adjusted for the product's administration fees. If the product's return comes in more than 0.5 percentage points below that tailored benchmark, it fails.

The design has a genuine elegance to it, and the elegance is the source of both the achievement and the problem. Because the benchmark uses the trustee's own asset allocation, the test does not punish a fund for being conservative or aggressive. A fund that chooses to hold 70 per cent in defensive assets is measured against a benchmark that also holds 70 per cent in defensive assets. The test is therefore not asking "did you make a lot of money?" It is asking a much narrower and much fairer question: "given the asset mix you chose, did you deliver what a cheap index-tracking version of your own strategy would have delivered?" It isolates skill and cost from luck and market direction. That is a defensible thing to measure, and it was a real advance on what came before.

What happens when a product fails

The consequences are automatic, public and severe by the standards of financial regulation. A trustee whose product fails must write to every affected member within 28 days, in a form largely prescribed by regulation, telling them the product has been determined to be underperforming and directing them to the ATO's comparison tool. APRA publishes the list of failing products. If a product fails the test in two consecutive years, the trustee is prohibited from accepting new members into it. In practice that is a commercial death sentence: a default product closed to new members has no future, and the trustee's realistic options narrow to fixing it fast or merging the fund into a larger one.

The test applied initially to MySuper products — the simple, low-cost default products that receive contributions when a member makes no choice at all. From 2023 it was extended to a much wider population of "trustee-directed products" in the choice segment, which had never been subjected to a standardised, published performance benchmark before.

The evidence: how a benchmark rewrites a trustee's incentives

This is the mechanism the entire article rests on, and it is almost never explained properly in public debate. It takes about six paragraphs. Once it is clear, most of the argument resolves itself.

The asymmetry at the centre of the test

Consider a trustee with two options that have identical expected returns. Option A is to hold the benchmark: the same asset classes, in roughly the same proportions, tracking the same indices as everybody else. Option B is to do something different — a large allocation to an asset the index does not contain, or does not contain in the same form.

Now think about what each possible outcome costs the trustee.

If the trustee holds the benchmark and the market does well, the fund does well, and nothing happens to the trustee. If the trustee holds the benchmark and the market does badly, the fund does badly — but so does every comparable fund, the benchmark falls with it, and the product still passes the test. Nothing happens to the trustee. Being ordinary is safe in both directions, because the test measures the product against a benchmark built from the product's own strategy, not against an absolute target.

If the trustee deviates and is right, the product beats its benchmark. The reward is a better number in a league table, some favourable coverage, and members who are mostly not watching. There is no formal recognition in the test framework. Passing is binary. A product that beats its benchmark by three percentage points and a product that misses it by 0.4 percentage points are treated identically by the legislation: both pass, both send no letter, both stay open.

If the trustee deviates and is wrong — even temporarily, even for reasons that have nothing to do with the quality of the underlying assets — the product can breach a half-percentage-point threshold. The trustee must then write to every member saying, in prescribed language, that their product has been determined to underperform. APRA publishes it. The financial press covers it. Members and employers move money. A second consecutive failure closes the product to new entrants, which in the default segment ends it. Boards face questions from members, from the regulator and often from a parliamentary committee.

The penalty for being different and wrong is public, legislated and potentially fatal. The reward for being different and right is nothing at all.

That asymmetry — a large, formal, asymmetric penalty for deviation and no formal reward for it — is what economists call benchmark hugging, or herding. It does not require anybody to behave badly. It does not require a single trustee to breach a duty. It is what a rational, conscientious, risk-aware board does when the rules make deviation expensive and conformity free. Any test with a hard failure threshold and no upside symmetry will produce it, and the effect strengthens the closer a product sits to the threshold and the more consequential failure becomes.

It is worth being precise about what this does and does not mean. It does not mean Australian funds hold no unlisted assets; they hold a great deal, and some of the largest are significant global investors in infrastructure and property. It does not mean the test has caused a measurable collapse in unlisted allocations, and no such claim is made here. What it means is that the test changes the price of a decision at the margin. It makes the marginal unlisted allocation — the next one, the contested one, the one that goes to a committee — harder to approve than it was, and it makes the argument against it stronger and easier to state than the argument for it. Marginal decisions, repeated across a $4 trillion pool over a decade, are how capital allocation actually gets determined.

Why unlisted assets are the hard case

The benchmark for each asset class is a listed market index — a published, daily-priced series tracking securities traded on an exchange. That is unavoidable if you want an objective, non-negotiable yardstick, because listed indices are the only returns series that exist independently of the entity being measured. Nobody can argue with the closing price of an index.

The difficulty is that many of the assets Australian funds hold, and most of the assets the country wants financed, are not listed anywhere. An unlisted toll road, a wind farm under construction, a stake in an unlisted technology company, a portfolio of build-to-rent apartments — none has a daily price, because none of them trades. They are valued periodically, typically quarterly, by independent valuers using discounted cash flows and comparable transactions.

Under the test, holdings like these have to be measured against the closest prescribed listed proxy. Unlisted infrastructure is benchmarked against a listed infrastructure index; unlisted property against a listed property index. This is a reasonable engineering compromise, and it is also the source of the whole dispute, because the proxy and the asset can move quite differently over multi-year periods. Listed infrastructure equities are shares in companies, so their prices carry equity market sentiment, gearing, interest rate expectations and the general mood of the exchange. A specific unlisted asset carries its own contracted revenues and its own construction and regulatory risk. Over ten years the two can diverge substantially, and the divergence has nothing to do with whether the trustee made a good investment. It is measurement noise that the trustee cannot hedge and cannot control, and under the test the trustee bears it.

The industry's peak bodies have made this argument formally and repeatedly. The Association of Superannuation Funds of Australia (ASFA) and the Financial Services Council (FSC) have argued in submissions to Treasury that the test's benchmark settings discourage investment in unlisted assets — infrastructure, renewable energy and venture capital in particular — because those assets cannot be sensibly benchmarked to listed indices. These are SECONDARY sources and they should be read as such: ASFA and the FSC are industry associations representing entities subject to the test, and they have a direct commercial interest in the outcome of the review. Their argument is not a neutral finding. It is a submission by an interested party, and its force comes from whether the underlying mechanism holds up, not from who is making it.

Valuation lag, and why it cuts both ways

The technical heart of the problem is timing, and it is the one part of this debate where the honest answer damages both sides.

Because unlisted assets are valued periodically rather than continuously, their reported values move after listed markets move, not with them. Industry submissions to Treasury put the typical lag for private equity at around three to four months (SECONDARY; industry submissions, Treasury consultation). When listed markets fall sharply in a quarter, an unlisted portfolio's reported value often does not fall until the next valuation cycle, and often falls by less. When listed markets surge, the unlisted portfolio lags upward too.

For a benchmark-relative test measured over defined windows, this creates real and unavoidable comparability problems. A fund with a large unlisted allocation can post a test result driven substantially by the timing of valuations rather than by the quality of its investing. Over a ten-year window the effect washes out considerably, which is one reason the window was lengthened — but "considerably" is not "entirely", and the residual noise sits right where the half-percentage-point threshold does.

Now the other edge of the blade, which the industry's critics press and which a fair account must state with equal force. Periodic valuation does not just create noise; it smooths reported returns. A portfolio whose value is marked four times a year by a valuer will show far less volatility than an identical portfolio marked every day by a market. Since risk-adjusted performance measures divide return by volatility, artificially smooth returns make an asset look better on a risk-adjusted basis than an equivalent listed exposure — not because it carries less risk, but because the risk is measured less often. Critics of large unlisted allocations argue, with justification, that this smoothing is itself a distortion, that it can flatter both reported performance and the apparent diversification benefit, and that valuation practices for unlisted assets involve genuine discretion exercised by the party whose performance is being measured.

So the honest position is uncomfortable for everyone. The test does impose an artificial penalty on unlisted assets through proxy mismatch. Unlisted assets do enjoy an artificial flattery through valuation smoothing. Both effects are real. Neither has been shown to cleanly offset the other, and no participant in this debate is a disinterested observer of the valuations.

The other half of Canberra is asking the same pool for the opposite thing

None of this would matter much if superannuation policy existed on its own. It does not. Across the same period the test has been operating, the Commonwealth has built an entire economic strategy around attracting long-horizon private capital into exactly the asset classes the test makes hardest to hold.

The Future Made in Australia package, announced in the 2024-25 Budget, committed around $22.7 billion over a decade to support domestic manufacturing and clean energy industries, on an explicit premise that public money would crowd in far larger private co-investment (Budget 2024-25, Australian Government). The National Housing Accord agreed in 2022 set a target of 1.2 million new well-located homes over five years from 1 July 2024, and named institutional investors — superannuation funds prominent among them — as intended participants in financing the supply. The Rewiring the Nation program committed $20 billion in concessional finance through the Clean Energy Finance Corporation to upgrade and expand transmission. And the Australian Energy Market Operator's Integrated System Plan sets out an electricity system transition requiring investment in generation, storage and transmission in the order of $100 billion or more over coming decades (approx.; AEMO, Integrated System Plan, 2024).

Every one of those programs points at the same money. Superannuation is the only pool in Australia with the scale, the duration and the liability profile to finance thirty-year assets, and ministers say so directly and often. Transmission lines, social and affordable housing, port and grid infrastructure, domestic venture capital: all of it is long-horizon, illiquid and unlisted.

They are also, without exception, the assets that sit worst inside a benchmark-relative test built on daily-priced listed indices.

The critique: two objectives, one pool, and nothing that reconciles them

The critique in this article is not that the performance test failed. It did not. It is not that the government is ignoring the problem. It is not — Treasury has acknowledged it in writing and consulted on it publicly, and that is set out below. The critique is narrower and, because it is narrower, harder to wave away.

The Commonwealth is pursuing two objectives that pull in opposite directions on the same asset base, and it has no mechanism that reconciles them. One sits with Treasury's retirement income policy and APRA's supervision: protect disengaged members by holding every product to an objective, non-negotiable benchmark. The other sits with the industry, energy and housing portfolios: mobilise the largest patient capital pool in the country into long-lived domestic assets. Both are legitimate and both are held sincerely. There is no forum, no committee and no statutory process whose job is to decide what happens where they collide.

Consider how differently these two objectives are enforced. The performance test is legislation. It has a threshold, an annual cycle, a regulator, a compulsory member letter and a product-closing consequence. It binds. The investment-mobilisation objective is a strategy, a set of budget measures, an occasional ministerial roundtable and a great deal of encouragement. It persuades. When a hard-edged legislated test meets a soft-edged encouragement, the test wins every time, and it wins inside a boardroom where no minister is present and no announcement is made.

There is no single document in which the Commonwealth weighs member protection against national investment capacity and states where the line falls. In its absence, the line is drawn by default — by whichever instrument has the sharper legal teeth.

The test's greatest achievement is invisible to the people it protects. Its cost may be invisible too — a decade of investments that were never made, and therefore never counted.

There is a second, subtler problem. The cost of the test's design is almost impossible to measure, because it consists of things that did not happen. A failed investment produces a number; a deferred investment produces a minute in a committee pack and nothing else. Australia has good data on what superannuation funds hold, and none at all on what they considered and declined. This is an asymmetry of evidence, not just of incentives, and it systematically favours the status quo: the benefits of the test are countable and published, and the costs are not.

The attribution here should be made precisely, because it cuts across the usual lines. The previous government caused this design: it drafted the benchmark architecture and legislated it in 2021. The current government inherited it, kept it, and has presided over four further test rounds. It also caused the other half of the tension, by building a signature economic agenda that depends on mobilising the very capital its inherited instrument discourages. Neither government designed the conflict deliberately. Both are responsible for a different half of it, and only one of them is now in a position to resolve it.

The counter-argument that has to be answered

An informed defender of the current settings would say the following, and it deserves a direct response rather than a rhetorical one.

First, that the ten-year window already handles most of the timing problem, because valuation lags of three to four months are immaterial across a decade. That is substantially right, and it is why the window was extended. The residual objection is that a ten-year window does not eliminate multi-year divergence between an unlisted asset and its listed proxy, and it is divergence rather than lag that does the damage near the threshold.

Second, that the test uses the trustee's own asset allocation, so a fund that wants to hold more unlisted assets simply has a benchmark that reflects that choice. This is the strongest defence available and it is genuinely powerful. The gap in it is that the benchmark reflects the asset class, not the asset: choosing a higher weight to infrastructure means being measured against a listed infrastructure index at that higher weight, which increases rather than reduces exposure to proxy mismatch.

Third, that no evidence establishes a causal link between the test and lower unlisted allocations. This is true and it must be conceded plainly. What exists is a well-documented mechanism, consistent submissions from the entities making the decisions, and Treasury's own acknowledgement that the concern is serious enough to consult on. That is a case for careful reform. It is not proof of a completed harm, and this article does not assert one.

What they got right, and the caveats

Three things belong on the record before anything further is said, and they should be said plainly rather than as a throat-clearing exercise before a critique.

The first is that the performance test worked at its primary job. It was designed to identify persistently underperforming default products and force them out of the system, and it did so quickly, transparently and at scale. Thirteen of 76 MySuper products failed the first test in August 2021 (APRA, 31 August 2021). Within a few rounds the failing cohort in MySuper had shrunk dramatically, and in the most recent rounds APRA has reported that no MySuper product failed. Underperforming products were repriced, restructured, closed or merged. Two decades of reviews had recommended acting on entrenched underperformance; this instrument acted on it in about three years. For members who never look, never switch and never complain, that is a large, concrete and durable win, and no argument about benchmark design diminishes it.

The second is that this government has not ignored the unlisted-asset problem. Treasury has acknowledged it publicly and consulted openly on it, through the Your Future, Your Super review and the subsequent Strengthening the Superannuation Performance Test consultation paper. That paper set out four areas of potential reform, including changes to benchmark settings for alternative assets, and canvassed the option of creating a new asset class benchmarked against consumer price inflation plus a margin rather than a listed index (Treasury consultation paper). Whatever one thinks of those options, they are a direct engagement with the mechanism described in this article, not a deflection of it.

The third is that the reform options published are serious rather than cosmetic. Reworking benchmark settings for alternatives, and contemplating a CPI-plus asset class, are substantive changes to the architecture of a test the government inherited and could easily have left alone. Governments regularly consult in ways designed to produce no change. This consultation put real options on the table, including options the funds asked for and options consumer advocates warned against. That is what a genuine attempt at a hard problem looks like.

The caveats on that credit are equally honest, and there are four.

Consultation is not resolution. A published set of options is the beginning of a policy process, not the end of one, and the tension described here persists in the meantime — in every investment committee meeting held while the review runs. Second, the options themselves carry real costs that the steel-man above sets out: a CPI-plus benchmark is easier to clear in a low-inflation decade, and every softening of a benchmark makes underperformance harder to detect and easier to dispute. Third, the test's success has made reform harder: an instrument with a demonstrated record of removing bad products is expensive to loosen, and the burden of proof sits on those proposing change — the correct place for it, but a heavy one. Fourth, nothing in the consultation addresses the coordination failure directly. Fixing the benchmark for alternatives would ease the friction at the margin. It would not create the missing mechanism by which the Commonwealth decides, in one place, how much member protection it is prepared to trade for national investment capacity — or states clearly that it is prepared to trade none.

It is also worth conceding a point against the framing of this entire article. It is possible that both objectives can be satisfied, that a well-designed benchmark for alternatives removes most of the deterrent while keeping the discipline, and that in five years this will look like a technical problem that was solved by technical means. That outcome is entirely available. The reason to write about it now is that it is not yet the outcome, and the decisions being deferred in the meantime are decisions about assets with thirty-year lives.