Australia has a productivity problem. You've heard this before. But the usual story - a gradual slowdown, structural shifts toward services, Baumol's disease - misses what's actually happened.
The real story is starker: Australia had exactly one period of meaningful productivity growth in the last 45 years. It lasted about twelve years. Before and after: stagnation. And since 2015, even the strategies that previously masked this problem have stopped working.
How economies grow
There are only three ways an economy can produce more output.
More hours worked. More people in jobs, or the same people working longer. This grows the economy but doesn't make anyone more productive - it's just more input producing more output.
More capital. Give the workers better tools - machines, software, buildings, equipment. This is "capital deepening," and it does raise output per hour. But it requires continuous investment, and eventually hits diminishing returns.
Greater efficiency. Find smarter ways to combine labour and capital - better technology, improved processes, more effective organisation. Economists call this multifactor productivity (MFP). It's the only source of growth that doesn't require ever-increasing inputs.
The first two have natural limits. You can't endlessly add workers or capital. Only the third - genuine productivity improvement - can sustainably raise living standards over time.
Australia's problem is that we've relied almost entirely on the first two, and never really achieved the third. Now capital deepening has stalled, leaving hours worked as the only source of growth. That's why GDP keeps rising while living standards don't.
One good decade in forty-five years
Start with multifactor productivity - the measure of genuine efficiency gains, stripped of the effects of simply adding more workers or more capital.
Since 1978, Australia's MFP has grown by just 18%. But even that modest figure flatters us. Almost all of it happened in a single window: 1992 to 2004.
Before 1992: flat. After 2004: flat. That twelve-year surge - driven by tariff reform, financial deregulation, competition policy, IT adoption, and the productivity dividend from the 1990s recession - was the exception, not the rule.
Labour productivity looks better, up 82% over the same period. But that growth came almost entirely from capital deepening: giving workers more equipment to work with. We bought productivity growth rather than earning it through innovation or efficiency.
The post-2015 stall
Here's what the standard narrative misses: even the capital deepening strategy has stalled.
Capital per hour worked rose roughly ninefold between 1978 and 2015. Then it stopped. For a decade now, capital deepening has been flat.
The COVID spike you see around 2020 was an artefact - hours worked collapsed while the capital stock didn't, mechanically pushing the ratio up. Once hours recovered, we were back to the same flat trend.
The capital-output ratio tells a similar story. After rising steadily for decades (meaning we needed ever more capital to produce each unit of output - a sign of diminishing returns), it too has flattened since the mid-2010s.
This isn't capital becoming more productive. It's investment failing to keep pace with employment growth.
The mechanism isn't mysterious: rapid labour supply growth from immigration reduced pressure on firms to substitute capital for labour, while accelerating depreciation raised the hurdle rate for net capital expansion. Why invest in automation when you can hire?
Where the growth is coming from
If capital deepening has stalled and MFP is flat, where is growth coming from?
Hours worked.
Total hours worked in the economy have more than doubled since 1978 and have continued rising sharply in recent years - largely thanks to immigration. More people working more hours produces more output. But it doesn't raise output per hour. It doesn't improve living standards for the people already here.
This is the uncomfortable arithmetic: capital deepening flat, MFP flat, and hours up. Real wages have grown - but only around 1.4% per year over the past decade, and even that modest gain hasn't come from productivity. Profit share has fallen sharply. Workers are claiming a larger slice of a stagnant pie. That's not sustainable: squeezed margins eventually mean less investment, fewer jobs, or both.
What about Baumol's disease?
A common explanation for productivity slowdowns is Baumol's cost disease: as economies shift toward services (healthcare, education, hospitality), productivity growth naturally slows because these sectors are inherently harder to automate or scale.
There's truth to this. But it doesn't explain what we're seeing.
Baumol's disease has been operating for decades. The sectoral shift toward services has been gradual and continuous. It can't explain a sudden stall after 2015.
And if high wages - which Australia has - were driving firms to substitute capital for labour, we'd expect to see capital deepening accelerating, not flattening. Firms facing expensive labour should be investing in automation. Instead: nothing.
Something else is going on.
(And before the economists object: yes, measurement error in services and intangibles is real. But it's been real for decades. It can't explain a post-2015 break either.)
The depreciation problem
One under-appreciated factor: the changing nature of capital itself.
Capital stock growth has fallen dramatically - from 8-9% annually in the 1960s to around 2.3% today.
Part of this is base effects (it's harder to grow a large stock at high percentage rates). But there's another dynamic at work: modern capital depreciates faster.
The IT revolution transformed the composition of business investment. Software, computers, and digital equipment now make up a larger share of the capital stock. Unlike buildings or heavy machinery, this equipment becomes obsolete quickly. A server from 2019 is ancient. Software requires constant updates. The technological frontier moves fast, and staying on it requires continuous reinvestment.
This means a larger share of gross investment goes to replacing depreciated capital rather than expanding the productive stock. You're running faster just to stay in place.
Why AI might not save us
This depreciation dynamic matters enormously for the AI optimism currently sweeping economic commentary.
The AI boom requires extraordinary capital investment in hardware that depreciates rapidly. Unlike previous technology waves, this creates a structural drag on productivity gains.
Consider the economics. A high-end AI chip costs tens of thousands of dollars and faces competitive obsolescence within 3-5 years as each hardware generation leapfrogs the last. Data centre operators must extract enormous value quickly just to cover capital costs - before accounting for power, cooling, and operations.
This differs fundamentally from the PC and internet revolutions. Software, once written, scales at near-zero marginal cost. The internet's productivity gains compounded as hardware costs fell and network effects grew. The marginal cost of sending an email or loading a webpage trends toward zero.
But AI inference is compute-intensive at the point of use. Every query, every generated image, every automated task consumes expensive silicon that is simultaneously depreciating. The technology doesn't get cheaper to run in the same way previous digital technologies did.
To be clear: AI may generate strong private returns for firms that deploy it successfully. The question is whether those gains show up at the macro level once capital costs are netted out - or whether we get another round of "computers everywhere except the productivity statistics."
If AI remains inference-intensive, hardware-dependent, and energy-constrained, then much of its gross productivity potential may be absorbed by accelerated capital depreciation rather than translating into broad-based productivity gains.
We may see impressive output from AI systems. We may see specific tasks transformed. But the aggregate productivity statistics could remain stubbornly flat as the gains are eaten by the capital costs of producing them.
What would actually help?
If the diagnosis is right - genuine efficiency gains stalled for decades, capital deepening exhausted, and AI unlikely to provide a free lunch - then the policy implications are uncomfortable.
The 1992-2004 productivity surge didn't happen by accident. It followed wrenching reforms: tariff cuts that exposed Australian industry to international competition, financial deregulation, national competition policy that opened up sheltered sectors, tax reform. These were politically costly and took years to bear fruit.
Nothing comparable has happened since. Instead, we've had two decades of relative policy stability (some would say complacency), punctuated by a mining boom that temporarily masked underlying weakness.
Genuinely lifting productivity would require similar ambition: serious competition policy reform in concentrated sectors, planning and zoning changes that let resources flow to their best uses, education and training systems that actually respond to labour market needs, tax settings that encourage productive investment over rent-seeking.
None of this is easy. Most of it is politically unattractive. The beneficiaries are diffuse, future and silent; the losers are concentrated, present and vocal.
The bottom line
Australia's productivity story is not a gradual decline that might reverse with the next technological wave. It's a single twelve-year surge, bookended by stagnation, with even the capital-deepening workaround now exhausted.
We've spent a decade with flat capital intensity, flat efficiency, and growth coming almost entirely from putting more people to work. That's kept headline GDP growing. It hasn't lifted living standards.
AI enthusiasm notwithstanding, there's no technological cavalry coming. The economics of inference-heavy, rapidly-depreciating AI capital may disappoint those expecting a productivity miracle.
If we want productivity growth, we'll have to earn it the hard way - through reforms that increase competition, improve resource allocation, and reward genuine innovation. The alternative is more of the same: headline growth powered by population, living standards going nowhere, and endless hand-wringing about a productivity problem we're unwilling to actually solve.
Data sources: Australian Bureau of Statistics (ABS) National Accounts (5206.0) and Modellers Database (1364.0.15.003). All series seasonally adjusted.
No comments:
Post a Comment